AI on Jamie Thingelstad

The nnxyz rooms

Thu, 15 Oct 2026 21:11:13 -0500

Years ago I discovered the work of nnxyz on nifty.ink. I acquired a few of his pieces and grew to like it more and more. I ended up collecting four complete collections of hundreds of pieces. I had always wished to create a gallery like experience for anyone to experience. With the help of AI I was able to make that vision a reality.

Check out the nnxyz rooms.

Some highlights of what the website has.

Every drawing can be watched being made. The 574 Nifty Ink works are rebuilt from their archived brush strokes as native vector images, and each one replays stroke by stroke, with the recorded order, widths, colors and transparency.
Hung like a museum, not a feed. Six wings, 28 rooms, 626 works. Every work hangs in exactly one room, rooms hold 20 to 30 works, and each room is hung in the order the works were drawn, so a walk through a room is a walk through a few weeks of drawing.
Wall text in a museum voice. Each room has an introduction, chapters, labels, one-line notes, explained comparisons and a closing.
An audio tour in every room. Press play and a voice reads the room while each work comes on screen, about seven minutes a room, with pause, skip and a progress line.
A screen saver that draws. One work at a time, drawn stroke by stroke on black, held with its caption, then the next. Fullscreen, with your choice of drawing speed and hold time, and it can be limited to one room.
A map of the building. The collection as a three-floor museum map, rooms sized by how many works they hold, colored by collection, with doors between consecutive rooms and stairs between floors.
Readable by machines too. /llms.txt indexes the site for language models and /llms-full.txt carries the complete gallery text in one file.
Light or dark. A white gallery by default, black by choice, remembered across visits. The artwork itself is never altered.
Everything downloadable. Each artwork page offers the native SVG, the original PNG and the archived metadata and provenance records, with a link to the collected NFT.

WT353 — Release Notes for Every Day

Sun, 11 Oct 2026 07:27:58 -0500

Claude Haiku 5.5, hard budget caps on usage-based APIs, zkAPI private payment credits, rumored Apple M8 Ultra AI servers, and Tufte’s Razor on the data-ink ratio, along with the launch of Your Version Number: Release Notes and the Twin Cities Marathon. Weekly Thing 353

Wed, 07 Oct 2026 18:11:44 -0500

Thingy, my digital librarian, got a big improvement this week with a brand new design and moving its reasoning to the 5.5 level models (Sonnet for subscribers, Opus for supporting members). I’m very happy with how this is working now!

Weekly Thing Archive Improved

Tue, 06 Oct 2026 21:30:00 -0500

Over the 10 years that I’ve published the Weekly Thing I have used three different platforms to send it. Each one of those had unique requirements for how I structured the issues. For a while I’ve wanted to get the entire archive updated so that no matter how far you go back the issues have the similar structure that they have today. With the help of Claude I was able to make this happen! The full archive is now updated renders cleanly. It required pretty extensive source doctoring behind the scenes.

Along with that I was able to then render every issue in audio as well. Now the Weekly Thing Podcast feed goes all the way back to the very first issue as well which is pretty cool.

mb 2.0 with MCP support

Mon, 05 Oct 2026 21:30:00 -0500

In March I created mb, a command-line micro.blog client that is specifically created for agents to use. This is what Otto uses to publish on its blog. It can also read the micro.blog timeline, follow people on micro.blog, and a variety of other features.

Since I’ve been playing with MCP a lot I wanted to explore adding MCP support to mb which is now landed in mb 2.0! In addition to MCP support I also created a Homebrew “tap” for it to make it super easy to install, added support for an mb --doctor command to help with setup, and a bunch more.

To try it the easiest thing is to use Homebrew:

brew install jthingelstad/tap/mb

Then sign in:

mb auth

mb auth walks you through signing in. It sends you to micro.blog’s app token page (Account, then Edit Apps) and helps you set the token. It checks the token with micro.blog, and if your account has more than one blog it asks which one to use. Run mb doctor afterward to confirm everything is set up correctly.

From there:

mb heartbeat  # what changed since last time
mb inbox. # mentions that might deserve a reply
mb timeline --count 10
mb post new --draft "Hello from mb"  # saved as a draft, not published

The default output is compact plain text tuned for agents. Add --human for something easier to read, or --format json for scripts. Use --draft or --dry-run to check a post before it goes live. Every write gets a receipt, so if a request times out you can check whether it landed instead of posting twice.

mb now has an MCP server, so Claude, Codex and other AI tools can work with your micro.blog account directly. To add it to Claude Code in read-only mode:

claude mcp add mb -- mb mcp --consumer claude-code --read-only

Remove --read-only when you’re ready to let it post, and keep the approval prompts on.

Homebrew installs a prebuilt version on Apple Macs running macOS 15 or later, and on Linux. Each release is tested against the real micro.blog API using a dedicated test account. If you’d rather skip Homebrew, the README covers installing with uv. Feedback and issues are welcome on GitHub.

Sun, 04 Oct 2026 21:17:36 -0500

I was inspired by the time analysis I had Claude do last week on what it took to send the Weekly Thing so I decided to build in some additional logging and a better visualization. Now I get a report like this for each issue I send. WT352 showed some benefits from my new automation! 🪄

Sun, 04 Oct 2026 07:02:07 -0500

Working with Otto this morning to create a new version of my mb utility that will extend features and provide an MCP interface for agents to interact with micro.blog.

Sat, 03 Oct 2026 12:00:00 -0500

My newest MCP experiment was to add it in WT Builder and then Claude was able to watch me creating content in near real-time through tool calls and give me immediate feedback on typos and grammar suggestions.

Wed, 30 Sep 2026 16:52:43 -0500

Oh no — I’ve used all my tokens! 😬 And I did successfully complete my mission to use the $250 of cloud credits.

Sat, 26 Sep 2026 08:44:28 -0500

I guess I’m getting my money’s worth for my $200 Claude Max account.

I went through the local Claude Code history on this Mac: 120 sessions and 254 subagent runs from Aug 6 to Sep 26, about 52k model calls. At API list prices that comes to about $13.7k. If you’re on a subscription, treat the dollar figures as a measure of how fast you use up your limits rather than a bill. Recent weeks run about $2.1–2.5k each.

There are obvious efficiency improvements for me to make. 😬

Fri, 25 Sep 2026 21:13:40 -0500

Easy mission: I will use all $250 of these free Claude “cloud session” tokens.

Weekly Thing Podcast Improvements

Mon, 21 Sep 2026 21:24:36 -0500

Claude and I did a session in WT Builder focused on the generated audio for Weekly Thing issues. This was one of the goals I had with the new authoring app.

Audio pauses will now fall in natural breaks in the flow. Each functional block of text is generated on its own.
Better section announcements and handoffs.
Blockquotes “quote” and “end quote” again.
Lists are spoken better.
I’m now embedding images from the issue into the metadata. If your podcast player supports that you will see them as the podcast plays.
Better pronunciation.
The descriptions now contain the links, so for the first time ever you can actually get to the articles from the podcast version.

With these changes the podcast version really can be a viable way to engage with the Weekly Thing.

Sun, 20 Sep 2026 16:38:31 -0500

I just published Weekly Thing 350 which is notably the very first issue to go through my new WT Builder app that I created with Claude. It worked pretty well and since it was first time on new code I told Claude to literally watch me use it on the server it was running on and fix things in realtime.

Building and Puzzling about Software

Thu, 17 Sep 2026 20:00:00 -0500

This originally started as a commentary on a link for the Weekly Thing and then got so long it became a blog post of its own. Many themes in this article start from my Software is Liquid article from April.

The craft of building software is in the largest state of disruption I’ve experienced in my 30+ year career. AI is changing everything. All the inputs and outputs are in flux. And the fundamental patterns we’ve worked on are all being redefined.

This isn’t the first time I’ve seen major disruption like this. My career starts with the web being born so I have no perspective before that, but I was part of the mobile and cloud transformations, and they profoundly impacted our craft. Releasing software thousands of times a month was an unlock to velocity that only the cloud could make happen.

Laurie Voss suggests that we are all product engineers now. I think he is on the right path here, and one of his assertions landed hard with me. The demand for software is infinite. You may nitpick with infinite, but roll with it for a bit. As an industry we have gotten faster and faster, but the demand has always grown even faster than that. Now with AI our ability to build software has increased radically. Along with that, not surprising at all to me, the opportunities to build have increased even more.

With “agentification” of work, developers have the potential to work on every function in a company. Every single part of it.

Said another way, the software is eating the world meme introduced by Marc Andreessen is going to be accelerated by AI, with the lower cost and faster speed of software creation.

I’ve made this argument before: look at the website of your dentist, your insurance company, your kid’s school, or literally any department of any government, and you’re looking at software that is terrible not because nobody knows how to build better software, but because the people who need it can’t afford to pay for better at current prices. Then think about all the things software hasn’t touched at all, which is most things. Every small business runs on a spreadsheet and a group chat and a person who remembers stuff.

That means there isn’t now and isn’t going to be a glut of software developers, and anything that looks like one right now is a temporary transitional state. The demand for software, at least inside my 10 year horizon, is for practical purposes infinite, or software developers wouldn’t be as highly paid as they are.

But the job of a “programmer” is about to get very, very different. So different that you might not even recognize it as “programming” anymore, while still being recognizably “software development”.

I’ve been working with Claude Code and Codex for months now building software of all kinds. Building my own agent, building fun websites for hobbies, building data products, building my own bespoke tools, building tools for other agents, even making a game!

Through that I’ve been learning what is still hard, what is now easy, what do we need more of, what are the pitfalls, where can you go off the rails? The craft is in flux. I think Voss captures what I’ve also come to believe our craft is becoming — Product Engineering.

I love the history he shared.

Then software went commercial and, especially, consumer-facing, and the translation job changed shape. Consumers don’t want to sit in requirements meetings: they just want to be handed a thing they like. So the person whose job was understanding what people wanted stopped being an analyst who interviewed the business and became a product manager who studied the market, a role borrowed more or less directly from Procter & Gamble’s brand managers by way of Intuit and then Microsoft, where a programmer named Jabe Blumenthal invented “program manager” in the late 1980s because Excel for the Mac needed somebody to own what it should do.

The function moved into Product, and Product got separated from engineering as a career, and for the last twenty-five years we’ve had two professions where there used to be one and a half. I bring this up because it means the job I’m describing isn’t a speculative new thing that we’d have to invent. It’s a thing we’ve had for sixty years under two names. My speculation is that it’s about to collapse back into one job.

In the beginning we built things. And then because of the limitations of the work we split that up into a bunch of different roles. Some people should talk to the users. Some people should create the interfaces. Some people shall write the code and work with the machines. Other people shall run it and deploy it. And yet others should test and validate it. I share those last two because they are roles that collapsed in many ways with the advent of DevOps. Role collapsing isn’t new in our craft.

Honestly it makes me chuckle a bit to sit back and reflect. My whole career has been undoing this fragmentation that started so long ago.

The web changed dramatically how user experience is created and understood. It needed to live in the product and be dynamic and move. You wanted to get signals from your users beyond formal feedback cycles.
DevOps and the cloud fundamentally changed the “over the wall” mentality of building and running software. Everyone runs the software.
Going back even more we used to have tons of specialization even on the build side. Much of that has collapsed as languages and infrastructure became more resilient, easier-to-use, and self-optimizing.

Now with coding agents we can create software at a pace that is reasonable for customers to engage with. We again have to collapse the walls. We can all engage in the product experience.

So what about programming?

Let’s look at this on a long scale. Go back to the 80’s or 90’s. Programming back then was about writing linked lists and managing memory. You were typically writing in low-level languages like C. Interpreted languages were ridiculously slow and lacked real power so we all lived by the law of the compiler. Memory was precious so you managed it precisely. Mind you, not as precisely as the machine language folks that preceded that era.

If you got in a time machine and showed those programmers the code folks are writing today they would laugh and say “that’s not programming!” You are just declaring a dictionary and shoving data in it? Where is that memory and how are you managing it? You don’t know? They would scoff at your lack of malloc, realloc, and free discipline.

Now we are back at that same spot but with agents. You don’t know how the functions are specifically implemented to make that feature work? You don’t know the precise database schema and how the thread pool is managed? Nope, but instead that engineer should and must be able to answer how context is managed, how the architecture is structured, how the event model works, and the most important of all — what is it we are trying to solve!

Personally I find the paradigm shift refreshing and invigorating. I can build and create things while ignoring a lot of the “how” that used to be required. For me personally, a lot of that “how” I couldn’t even do, so it wasn’t that it slowed me down, it was impossible for me. This is why AI makes your “impossible list” possible.

But for the folks doing the coding, I’ve seen different approaches based on two things people love to do: building and puzzling. All developers do both, but often identify their impact more strongly with one or the other.

The builders I know are loving AI and having a blast creating things. Builders have tended to be the folks that love to work on a specific domain and create something new. They love the act of building and creating something from nothing. Some builders (I fall in this camp) love to build so much that they really don’t even care that anyone uses it. It is almost like art — you build because it is fundamentally rewarding to create new things. Builders have taken to AI fast because it allows you to build so much more.

Puzzlers are the developers that we have always given the hardest problems to. Puzzlers can work on a problem for hours, days, weeks and can never put it down. These are the folks that achieve flow states where time stops and the only thing that exists is to find the next step in the puzzle. They feel rewarded because the puzzle was solved. These folks thrive on the hard, the complex, the esoteric. And technology teams need them badly because to get much of anything to scale you need puzzlers. The puzzlers I know are less excited about AI. It takes on work that is closer to what they love and value.

The funny thing to me is that we’ve had builders and puzzlers working hand-in-hand on the same projects for decades and they may not even know themselves the difference. The craft is the same after all. The difference is the reward cycle.

I do think puzzlers will also find the AI transformation rewarding but it will come in solving massively bigger puzzles, or perhaps solving dozens of puzzles at the same time. Here the cycle is less about the fast build, and more about the expansive data analysis. I was doing work on a large data set and it was so amazing to have Claude build several models, run them all for real, and see the variances. Supercharged puzzling! I think the puzzlers may be the best folks we have to build software factories today.

Where does this leave us

Voss suggests that we have “ten years of turmoil” ahead and I think he is likely right but I might put a different flavor on it.

Some things I think we’ll see:

The entry-level path into software development or product engineering will need to be redefined. The reason senior developers are so highly sought after is that they have the expertise and experience to make this transition now. We will need to redefine how people get started in this career. This has happened many times in the past. Each major technology transformation has disrupted the career ladder and then it re-stabilizes on the other side. That is cold comfort for someone hit by that transition now, but my main suggestion is to focus heavily on where the trend is going and get there as fast as possible.

I think a lot of people are going to build really terrible software, and the pain is going to be real. Just because you can use an agent to make software doesn’t mean you should. Running software isn’t a one-time job. And however you build it most software that starts life as a cute puppy eventually becomes a big smelly dog that you have to pick up after. Folks that are building their own systems with no expertise will not realize when they are already in a bad spot, and then something will fail or a security event will happen and it will hurt badly.

Lastly I do expect we will see an even greater unification of functions back to “creating products”. The specialization that pace and complexity introduced has been a complicating and challenging issue for decades. Ask any product team and development team to talk about the other one. They have all sorts of opinions about how the other could do their job better. The reality is they are all doing the same job — attempting to build amazing solutions for people! We’ve gotten confused as an industry that these are two things not one. It will be great for everyone to see this fade away.

All in it is an amazing time to be in this craft, and also one of the hardest times. I’ve never felt the need to learn so much, the fire to create so much, and the excitement of what we can make happen.

Sat, 05 Sep 2026 16:59:20 -0500

I’ve had Claude working on a new project and since I have the tokens I’ve used Fable 5.0 for nearly all of it. It has done an incredibly good job with a very complex system.

Thu, 03 Sep 2026 21:35:08 -0500

I’m ready to call Thingy complete! Thingy is the agent that has read everything I’ve ever published and helps you explore it. I recently added a ton of great stuff, including support for photos and sharing conversations. Check out exploring Ireland and going to Iceland with Thingy. Pretty cool!

Thu, 03 Sep 2026 20:57:50 -0500

Otto upgraded to OpenClaw 2. I had a super rocky start with OpenClaw but it has been working really well for a while now.

Thu, 03 Sep 2026 20:24:52 -0500

I had an idea to “flip” Elixir, the Clash Royale agent I created, upside-down and open up MCP use — Elixir MCP! Every single part of this I’ve created in other solutions. Claude (Fable 5.0) and I spec’d it out and I put Claude Code in a /loop and six hours later we had a working system.

Thu, 23 Jul 2026 13:36:51 -0500

From The Economist’s World in Brief today:

OpenAI said that GPT‑5.6 Sol, a frontier model, and another unreleased model hacked into a software company’s system without being told to do so. The lab said the models escaped its testing environment, went online and stole login credentials for Hugging Face, an AI platform. Hugging Face said that it used a Chinese model to defend itself because the guardrails on American models made them useless.

Wild.

Sun, 05 Jul 2026 15:01:26 -0500

Turns out ChatGPT is a fan of POAP KINGS! Wonder if our agent optimized interface is helping?

Elixir's Agentic Product Team

Mon, 22 Jun 2026 21:30:00 -0500

I’ve been exploring and engaging in agentic software deeply for a couple of months now with a number of my own projects to learn with. Thus far, nearly all of what I’ve done has been driven by me often with an LLM assisting in the framing of the thing I’m looking to build. I decided to try using “agents” (of a sort) to create an entire execution loop to autonomously drive one of these projects. I decided to focus on Elixir, the agent I’ve created to run our Clash Royale clan, and have now created six discrete automations for it.

Here is a brief overview of what each does, along with a link to the full instructions they use. I created the instructions for each using Claude Cowork and this is the AI-generated summary of each entity.

Data Analyst — Watches the Clash Royale API data, raw payloads, event streams, detections, and battle telemetry to find new data patterns, game changes, schema drift, unused data, and capability that the Product Manager should consider. Data Analyst definition.
Product Manager — Turns evidence into direction: reviews clan needs, leader feedback, quality reports, Discord history, RoyaleAPI content for editorial context, and data briefs, then files proposed issues for improvements that drive Elixir’s mission and await approval before build work begins. Product Manager definition
Build Manager — Converts approved, ready issues into small, tested code; owns feature and bug-fix implementation, respects issue scope, runs tests/evals, and hands deploy/restart needs to Operations. Build Manager definition.
Quality Manager — Judges whether Elixir is actually working: checks recommendation accuracy, silence/noise problems, routing failures, prompt failures, leader/member feedback, and regressions, then files actionable bugs, regressions, quality issues, or eval requests. Quality Manager definition.
Operations Manager — Owns production health: monitors runtime status, logs, telemetry, Event Core health, costs, retries, scheduled jobs, and delivery systems; fixes operational/reliability issues and handles deploys or restarts when needed. Operations Manager definition.
Evaluator — Owns measurement: builds and maintains eval harnesses, datasets, scoring rules, benchmarks, and regression tests so the team can tell whether changes improve or degrade Elixir with evidence instead of vibes. Evaluator definition.

The common interface that all of this runs through are Elixir’s Github Issues. All agents interface there and that is where “human in the loop” is happening before moving forward with changes. The automations are all setup in Codex, which is super simple and easy to engage with. I’m super curious to see where this goes!

Sun, 21 Jun 2026 07:56:02 -0500

My Claude Code + Codex 30-day token usage versus my subscription cost looks good. Just about 2B tokens. I suspect this is going to go up even more as I’m venturing into autonomous coding agents.

Sun, 21 Jun 2026 07:40:44 -0500

Just told Claude Code “you have broad approval to do whatever is needed to execute this effort” and then made me double-take about “super intelligence” risks. 😅

Fri, 19 Jun 2026 17:25:09 -0500

We’ll see how this goes…

I asked Codex to watch one of my agents, Elixir, and using suggestions from Elixir and its own observations automatically modify Elixir as needed with a GitHub Issue as a tracker.

I’ll be curious to see what they do together.

Mon, 15 Jun 2026 21:25:08 -0500

With some trepidation I’m taking my third try with Obsidian. This time with the promise of using it in collaboration with various agents. We’ll see how this works. Obsidian being primarily used by Agents may actually be a worthwhile path.

Wed, 10 Jun 2026 06:46:19 -0500

I’m asking Fable 5 to look at some of the projects I’ve been working on and so far I’m very impressed with the level of analysis and recommendations it can make. Opus (via Claude Code) and GPT 5.5 (via Codex) were not able to see the structural issues in the code that Fable can.

Thingy Evolved

Sun, 07 Jun 2026 07:36:27 -0500

I rebuilt Thingy into the archive agent I always wanted. Thingy started a month ago as an agent that acted as a librarian in front of the Weekly Thing archive. It did a great job at that and even if nobody else used it I found it super helpful to explore the archive. However, I quickly found that I also wanted Thingy to know about the 12,000+ posts on my blog in addition to the Weekly Thing. Oh, and my one episode Another Thing podcast should be available too. I wanted Thingy to know all of the information I have published.

Core to making this happen was to create semantic archives for not just the Weekly Thing, but also my blog and Another Thing. Once I had that foundation, I gave Thingy super powers!

The changes…

A New Home: Thingy is now its own dedicated website at thingy.thingelstad.com. That gave it room to become a full chat experience, with a cleaner interface, mobile-first interaction, conversation history, and a user experience that feels like Thingy rather than a search box.
Search Box to Agent: The biggest conceptual shift is that Thingy no longer feels like you are querying a database. It welcomes you, remembers prior conversations, understands which archives are available, and can help connect threads across the Weekly Thing, thingelstad.com, and Another Thing.
Three Archives, One Conversation: Thingy now works across all three semantic archives: the newsletter, the blog, and the podcast. The source selector lets you focus or broaden the conversation, and the backend has been improved so each archive has similar and robust capabilities.
Persistent Conversations: Conversations moved from browser local storage to the server. That means Thingy can remember conversations across devices, show recent chats, reload prior threads, and provide robust evals as well as user-friendly naming.
Curiosity Maps!: This is probably the “show, don’t tell” feature. Instead of only answering questions, Thingy can draw a map of adjacent ideas from the archive. You can start with Thingy’s suggested map or seed it with your own topic.
Richer Responses: Responses can now render tables, trails, maps, citations, action buttons, copy/share/play controls, and better markdown. This makes Thingy feel less like a chatbot bolted onto an archive and more like an interface for wandering through ideas. Thingy also shares its thinking process with you as it works!
Voice Input: You can talk to Thingy now. It is not a full voice conversation mode, but speech-to-prompt makes the experience feel more natural, especially on mobile.
Mobile Experience: The mobile interface got a major rethink: chat-style header, conversation drawer, floating composer, cleaner controls, and a more app-like feel. This matters because Thingy now feels usable on a phone, not merely responsive.
Memory and Personalization: Thingy can remember user preferences like name and use prior conversation context. It can welcome returning users differently from new ones, and the experience can become more personal.
Thingy Identity: The UI moved away from a publication aesthetic and toward a dedicated chat client with Thingy’s own visual presence. The larger rail image, cleaner composer, and simplified interaction model make it feel like its own thing.

Thingy requires you to be subscribed to the Weekly Thing, and it will do the subscribing for you as you sign in. Give it a try! 🤩

Sat, 23 May 2026 08:50:52 -0500

A pattern I’ve adopted when I’m heavily engaged with Claude Code or Codex is to have an open text file where I’m keeping the “next prompts” that I’m considering. While the agent is working I’m continuing to refine these so we are ready to move to the next feature or task.

Fri, 15 May 2026 15:18:46 -0500

From a chat with Claude today…

“everything else is scope creep dressed as ambition”

Statistical word model or not, that was an insightful comment.

The Weekly Thing Team

Tue, 12 May 2026 21:00:00 -0500

I’ve been publishing the Weekly Thing for nine years and automation is one of the things that has made that possible. I shared a while back how I find content, assemble the issues, and my project structure. Without these well defined workflows there is no way I could continue this project.

The structure I have has worked well but it isn’t autonomous. It only runs when I engage with it. It is also brittle and “one way”. I can only easily run it one time. Additionally I think I could use some help getting things collected and reviewing the in process writing.

To this end I decided to create my support team for the Weekly Thing!

My starting points were:

Based on Elixir I know that Discord is a reasonable place for an agent to run.
Based on building Thingy I know that my Weekly Thing archive is a robust knowledge base to build off of.
Based on sending issues that meander and are just too long sometimes I know an editor would be helpful.
Based on my own time crunch that I get into when I’m trying to make a whole issue happen in one Saturday morning I know I could use some help making it more iterative.

This is the genesis of my Workshop and the four agent team that I have now created to assist me.

One thing worth being clear about: I have stated many times that “My words are mine!” and not AI’s and that is still the case. I don’t have any of these agents working to write content for me. They are my support team. The words are still mine. The only case where an LLM is “writing” or engaging with anyone is Thingy, the librarian for the Weekly Thing, and the Supporting Membership program where I have an explicit preference that that be a different voice than mine.

Here is the broad outline of the multi-agent solution that allows me to have dedicated agents that focus on different aspects of publishing the newsletter each week. This allows me to focus more on writing and commentary!

Each of these agents are operating with a full set of tools that include the entire archive of the Weekly Thing. As a result they are much more tuned to the job at hand than a generic LLM.

Eddy is my editor who reviews everything that goes in the newsletter. Eddy assembles a working draft of the current issue of the newsletter every day and then does an editorial review of the content. Eddy shares a status and progress indicator with me in Discord.
Linky is my researcher who assists with assessing the links I flag to go into the issue. Linky does recon to allow me to filter faster. Linky doesn’t ever look at the current issue and is just assisting with curation. Linky shares these in Discord. I’ve made it so I can reply to Linky with my commentary and it sync’s it back to Pinboard. This has allowed me to turn my commentary into a conversation.
Marky focuses on the most recent issue of the Weekly Thing that has been published and raising awareness. I’ve done the least with Marky so far, but the goal is to get the Weekly Thing to new readers.
Patty is the supporting membership manager who helps create call to action to bring new members in and raise money for the nonprofit we have selected. Patty operates on the annual cycle of the membership program and is the only agent that will draft content that does appear (properly sectioned) in the Weekly Thing. Patty understands the goal of the program, the organization that we are focused on this year, and what I have been writing about.

I’m focused mostly on Eddy and Linky right now as they are core to my authoring cycle. I can already see that this is going to allow me to focus on the content more, will be a quality of life improvement to get more incremental content and less scrambling at the end of the publishing cycle, as well as a more readable final email to subscribers.

Tue, 12 May 2026 17:00:00 -0500

I was ecstatic to share with TeamSPS that we now have an enterprise agreement for Claude. It was fun to get to share the stage with Erica Koenig to make the announcement noting the incredible capabilities we are putting into peoples hands. This is just the beginning.

Massive Weekly Thing Website Build

Thu, 07 May 2026 22:00:00 -0500

I’ve commented that agentic coding makes things that were previously on your “list of impossible projects” into things that you can do. I have long had on my “impossible project” list the desire to create a website for the Weekly Thing that let the archive shine in ways that I knew were possible but no solution out there delivered. With 9 years of writing and 345 issues in the archive there is so much to surface.

To do this I knew I would need to build it on my own. I could use the Buttondown API to get the issues and make them accessible. But then I needed a website. I needed a content pipeline. Oh, and that archive has old formats from different platforms that were a mangled mess of HTML.

This was truly on the “impossible list” for me personally. If I wanted to spend tens-of-thousands of dollars, or probably even more, I maybe could have hired someone to build it. A laughable idea really.

So I decided to take my experiences with Claude Code, Claude Design, and Codex and point it at this problem. Over the last couple of weeks I’ve been working on the new Weekly Thing website experience.

I just have to say I’m so thrilled with the results that I can barely handle it. Rather than type a novel here I’m just going to list out what the site has. Even better, go there and explore:

https://weekly.thingelstad.com

Here is what the new site has!

1️⃣ Completely reimagined landing page to describe the Weekly Thing. Gone is the basic Buttondown paragraph of text and a signup button. The home page hopefully gives a much better feel for what the Weekly Thing is.

2️⃣ Archive page has full index of every issue back to number 1. This is also now optimized for the Weekly Thing with issue images, link counts, organized by year.

3️⃣ Thingy, the Weekly Thing librarian that has read every issue of the Weekly Thing and is ready to converse with you about all of it. I have wanted to make an agent like this for over a year and it is finally real. I’ve found this fascinating to play with and ask questions of.

You will see this feature requires you to provide your subscriber email address. It is only available to confirmed subscribers of the Weekly Thing.

You may recall in WT311 I shared a custom GPT that was sort of like this. That was grade school level. Thingy is much smarter!

Some prompts that are fun to explore with Thingy:

4️⃣ Search is now super powered. The searching is indexed into the section of the weekly thing. This works way better than before.

5️⃣ On the page for each issue you will see that there is a Table of Contents on the left. It is a little thing, but another example of something I’ve wanted for a long time. The Weekly Thing is long and this gives a way to navigate. Also, each of those items is a hyperlink so you can now send a link to a specific notable link in a specific issue.

6️⃣ Big one – you can now LISTEN to the Weekly Thing. I’ve filled this in for the last 10 issues. On the issue page there is a “Listen” button where it will be read for you.

7️⃣ Podcast? Well if I have an audio file for each issue why not bundle that into a podcast. So I did. You should be able to find the Weekly Thing on Apple Podcasts and Spotify. It is propagating through other platforms. Should be on Overcast too.

8️⃣ Support for LLMs.txt! This is a bit hidden, but if you want to talk with the LLM of your choice about the Weekly Thing, give the LLM this link:

https://weekly.thingelstad.com/llms.txt

That provides an LLM optimized index of the entire 345 issues, as well as links to LLM optimized versions of every email! This means ChatGPT or Claude or whatever else can dive deep into the content. I have actually used this myself when asking a model to do some research with me.

A quick note about the audio:

This doesn’t replace or remove my actual podcast, Another Thing. There is still just one episode there but I’m not giving up on that.
The audio for the Weekly Thing is text-to-speech using a transformed version of the email text. It announces sections, gives links numbers, announces quotes, and cuts some sections. I’ve listened to a few and think it works reasonably well.
I’ll probably evolve the generated audio, and right now it only exists for the last 10 issues, but I plan to backfill all issues with audio over time.

Take a look. Try out the archive, search, Thingy. Listen to an issue. And let me know what you think… anything not work right? Read wrong? Something missing? Or just that you think it is all cool?

Your Version Number: The Website

Wed, 06 May 2026 22:00:00 -0500

In 2018 I wrote about Your Version Number:

I think the version metaphor works. You are a different person in your 20s, 30s, 40s and so on. Your life changes in meaningful ways! MAJOR version! Each year we tend to think of new things and new goals, but we don’t break backwards compatibility. MINOR version! And I think most people try to make each day a bit better than the last. PATCH level!

I’ve had a shortcut that shows my version on my phone for the last year. And now I decided to make this a fun website!

Now available…

Your Version Number

Super simple single-page app that allows you to add one or more birthdays and see the version number. The magic here is all the data is held in the URL so you can bookmark, set as your homepage, or share with others. And sure it is fun seeing you and your friends daily versions, but it is even more fun with all the custom themes!

dark — Minimalist dark mode with violet accents.
family — Warm cream + handwritten Caveat, family-album feel.
pastel — Soft gradient haze with pastel cards.
birthday — Confetti, balloons, and party-hat pink.
nature — Scattered leaves on linen, earthy serif.
ocean — Wavy gradients with a tiny shoreline wave.
galaxy — Deep-space gradient with neon numerals.
zen — Quiet cream with a vermillion first-letter.
weather — Sky-blue card with a sun/cloud per row.
polaroid — Taped Polaroid grid with a slight tilt.
tarot — Purple stars and Fool / Priestess / Empress cards.
newspaper — Broadsheet typography with section rules.
subway — Black NYC subway map with route bullets.
receipt — Thermal-printer monospace with QTY 1.
steampunk — Sepia gear-and-cog ledger.
brutalist — Yellow + red + black, oversized type.
comic — Comic panels with POW / ZAP / BOOM stickers.
memphis — 80s squiggles, triangles, and dots.
vinyl — Spinning 33⅓ records.
terminal — Green-on-black CLI prompt.
arcade — Pixel-fonted hi-score CRT cabinet.
vaporwave — Pink/cyan grid with a palm-tree sunset.
y2k — Frosted-glass chrome and blur.
pixel — Eight-bit pixel font on a dark green field.
gameboy — Classic GB DMG palette and cart silhouette.

Wait though, there is more.

Your Version Number: Work Edition

I decided to bring the same version number semantics to your job! Here we have YEARS.QUARTERS.BUSINESS_DAYS! Add your whole team in and share the URL with everyone.

And of course the Work Edition has work themes!

boardroom — Board-update slide with KPI rail (ARR, NPS, payback).
slack — Slack channel feed with reactions and avatars.
slidedeck — Confidential business-review slide with three bullets.
earnings — Live stock-ticker on a black trading screen.
github — GitHub PR list with avatars, labels, and the Open pill.
whiteboard — Sticky-note grid in primary colors.
inbox — Gmail-style inbox with subject lines and senders.
okr — Q-scorecard with progress bar and ON-TRACK pill.
cubicle — Manila-folder corporate newsletter.
kanban — Jira-style cards in an In-Progress column.
standup — Daily-standup card with Yesterday / Today / Blockers.
invite — Calendar invite with Accepted check.
confluence — Wiki page with breadcrumbs and comment counts.
zoom — Gallery-view tiles with reactions.
spreadsheet — Excel grid with row numbers.
pomodoro — Tomato-timer Deep Work card.
ooo — Out-of-office auto-reply with handwritten signature.

A fun little project made possible with Claude Code and I. 🤩

Wed, 06 May 2026 21:26:34 -0500

Prompt I gave Claude Code on a recent project.

I would like to iterate faster. I would like you to use the ANTHROPIC API key and ask Haiku to generate 20 questions for each agent persona that Jamie (me) would reasonably ask. Then run those questions to each agent and see how they respond. Adjust based on that and keep iterating until all 4 agents respond to the 20 questions okay. Make sense?

Worked well.

Sun, 03 May 2026 06:06:28 -0500

Up and writing at 6am for the next Weekly Thing!

While I’m doing that I’ve got Claude Code backfilling audio versions of the Weekly Thing, and another Claude Code instance doing some incremental work on Elixir.

Sat, 02 May 2026 15:00:00 -0500

Tyler led his first session at Minnebar today! He did an amazing job talking about AI in Schools and what is good, bad, and how schools should adapt. 👏

Sat, 25 Apr 2026 09:05:16 -0500

Refining Elixir’s agent definitions by having Opus review prompts that Sonnet has modified from OpenClaw’s refactor to be used by Haiku. Seems like magic. 🪄

Hypergrowth

Wed, 22 Apr 2026 18:00:00 -0500

This growth is hard to even comprehend.

Anthropic says Claude Code is now growing revenues at a $2.5 billion run-rate, a number that has doubled since January 1. Claude Code was launched in May 2025. Six months later it was at $1 billion. Its sales are growing faster than a 1980s F1 monster, pulling the whole company along with it. Anthropic hit $14 billion in ARR in February, $19 billion in March, and around $30 billion this month. — Software Eats Its Own

Even being prepared for that level of scaling is impressive.

Software Is Liquid

Mon, 20 Apr 2026 19:52:27 -0500

This started as a talk I gave internally to a group of technology leaders. I’ve adapted it here, stripping out the company-specific material, because the core ideas apply well beyond any one organization.

I want to throw out some ideas about what I think is changing in our industry. What I’m going to describe is one of the most rapidly evolving, most dynamic changes I have ever seen in a twenty-plus-year career in technology. I believe there are things changing right now that will fundamentally redefine how we practice our craft.

Where we are

Let’s bookmark where we are. Agents are real. I’ve watched one go from nothing – zero, no code, no design – to a working alpha with real users making real decisions on it, in about five weeks. A year ago, this was an idea. Now there are production agents running.

Agentification is the next major milestone our industry is going through.

I’m old enough to say this: there was a time before the web and a time after. A time before mobile and after. A time before the cloud and after. And now, a time before AI and agents and after.

I think this is going to be the most transformative of all of those.

You have to acknowledge one paradox of where we are: we are building the best practices before they exist.

We’ve been here before. Those of us who lived through the cloud transformation remember being ahead of the industry, figuring out how this ephemeral compute stuff works, how to make it all function. There weren’t patterns the industry had settled on. That’s where we are today. There are no clear patterns for how agentification happens. We’re going to build those patterns and learn alongside the industry.

That’s okay. Just be aware of where you are. It’s fine to be out ahead of the curve; you just have to always know when you’re there, because it’s a risky spot. You don’t want to be too far out.

Getting philosophical: software is becoming liquid

I want to get a little philosophical.

For the last two months, I’ve pushed myself into a level of AI engagement that is probably unhealthy long-term, honestly. If you ask my family, they would agree. But what I’ve been trying to do is really wrap my head around the core concepts that I think change how we do what we do.

Connect this back to other transformations. When we adopted a mobile world, we all knew we needed to be mobile users to lead it. Can anybody build a great mobile app if they’ve never used a mobile phone? Obviously not. So to lead through AI agentification, we have to be really close to it. I’ve been pushing myself hard to do that, because as you build experience, it gets harder to refactor how you think.

Here’s something I’ve been thinking a lot about: what is the cost of being wrong? And how do we fold that into how we create things?

Step back for a moment and think about how we do our craft. Say we’re building software. We spend time on discovery. We create stories. We have designers go off and make wireframes. We do all kinds of things to make sure that, when we actually get to the point of building, we know we’re doing the right thing.

Why? Because the act of building software has been incredibly expensive. The last twenty years of my career have been about figuring out how to effectively turn ideas into working software and how to make sure that, when we do, we’re not wrong – that we’re producing valuable capability. That’s what technology teams around the world have been focused on. The teams that do it well do these things better than the teams that don’t.

Here’s how I’ve come to think about it: software has historically been a solid. It’s chiseling something out of granite. We have our ideas, we sit down, it’s hard work, it’s challenging, and we chisel it out of granite.

I think that’s changing. I first heard this from somebody online and it didn’t land for me at first – I thought, that doesn’t make sense. But the more I thought about it, the more I thought it was spot on. The assertion was: software is becoming liquid.

We’ve operated for two decades in a world where software is an incredibly difficult solid to shape. With AI and agentic development – automatic programming – software is becoming malleable. If you’ve worked with agents on software, you’ve had the experience of thinking: I can refactor this code faster than it would have taken me to do all the guardrail work to make sure I didn’t make the wrong decision in the first place.

When the economics of what you do change that profoundly, you have to question everything.

Every paradigm built on “software is expensive” needs re-examination

Go back further. Most of us spent the early parts of our careers in agile flows. Before that, everybody did waterfall. Why? Were they just not as smart? No – they were operating under a different set of assertions. If you made a mistake writing C code, it was really difficult to unwind. You’d take months to refactor a mistake in your domain model that showed up in C code.

Then we got Python, PHP, interpreted environments, continuous integration. The paradigm changed. Suddenly: what if I’m wrong? Fine, I’ll refactor. Refactoring Python is cheaper than refactoring C. That’s just a fact.

So here comes agile. We can do this differently. We can be more responsive.

The cloud is the second part of that story. The cloud says we can do the same thing with hardware – we don’t have to worry about where we put the server. The cost of being wrong, if I put a server in the wrong data center, is not easy to undo. But in the cloud, that’s a couple of commands.

We are in that same spot again.

AI is transforming the cost of creating software in a way that should make us question every single process we have that is fundamentally built on the assertion that creating software is expensive.

I’d argue that maybe proof of concept doesn’t make sense anymore. What we used to call a proof of concept is now discovery. And how do you do discovery? I think you do it in code. Your discovery process is entirely in code. Do you then throw the code away? No. Why would I? It’s liquid. I bend it, move it around, get it where I want it.

The act of discovering through creating changes things pretty fundamentally. That paradigm is going to take us a while to absorb. I really want you to think about what you have in your world that makes assertions that building this stuff is extremely expensive. And when I say “cost,” don’t just think dollars – think organizational cost.

That’s assertion one: software is becoming liquid.

Managers belong in the code

Here’s my second assertion. If you’re a people leader, this one is for you.

I used to have a very firm belief. When I saw a people leader – a director or a manager – in code, that was a warning sign. Almost always, when I saw a director or manager in code, they were probably avoiding something harder that they were supposed to be doing. “Oh, you’re working on the actual software? I bet you have a personnel problem you’re not dealing with.”

I don’t think that’s true anymore. Agentic engineering changes that fundamentally.

The issue historically was a simple one of context window. As a manager, you couldn’t truly know the codebase because it was too complicated. It was a solid asset your craftspeople were working on. You had to focus on the people systems. You just couldn’t hold both of those things in your head and be effective.

Agents change that paradigm entirely. There is no reason, as a director or a manager, why you shouldn’t be talking to an agent and asking about the quality of the asset you’re accountable for. And as a people leader, you are accountable for the assets your people create. So why aren’t you having that conversation?

Why would I ever start a conversation with an engineer with, “how long do you think that’ll take?” I should have had that conversation with Claude Code first – looked at the source, asked: is this a big refactor? If we went this direction, how would that look?

This flips even further on its head when we think about agentification, because increasingly we’re going to be creating software not for people to use, but for agents to use. Think about how that works. You work with an agent to create the software. Another agent uses it. The agent using it gives you feedback on how it’s working. You take that feedback back to the coding agent and ask it to iterate.

What are you in that loop? I don’t know – a product manager, I guess.

I’m doing this today on multiple projects at home – using agents to give each other feedback. This speed and paradigm shift is foundational to how we have to adjust our thinking.

Rethinking velocity

The last thing I want you to really think about: as an industry, we need a step-function change in how we think about velocity.

How long is something going to take? I’d argue every paradigm you have for answering that question is broken now. The cost understanding is broken. The complexity understanding is broken. It’s all broken.

The only way to truly gauge it is through the second thing I mentioned – getting closer to the asset you’re accountable for, getting closer to the code and the product.

Just like mobile – where you couldn’t understand how to build an app until you’d experienced one – you can’t understand agentic transformation until you’ve experienced it firsthand.

Don’t be scared to go close to the code. Don’t be scared to ask your team, “Hey, how do I get that code out of Git? I’d like to look at it and do some analysis.”

These are superpowers. Every single one of us can put a cape on. You didn’t have these before. I think it’s amazing. And the whole industry is going to go through this transformation.

The change curve, and the rate of change

I want to close with something about change itself.

There’s a model called the Satir change curve. Every one of us sits at a different point on it right now. But just like every transformation before it, your progress is gated by your own engagement – by your own willingness to rethink the craft you have and to let go of things that may have been important for the last two decades but aren’t important anymore.

I invite you to come down this path.

Personally, it’s not easy. And what’s not easy is the rate. Think about it: we had two or three or four years to figure out mobile. We had half a decade for cloud. The web took an eternity – it was the first one. Here, we’re trying to do this in about a year.

Why? Because it’s enabled by all the other capability we’ve built, and because the potential is so big. The return on investment, once we identify things, is measured in weeks or months – not years. That’s a completely different thing than any of these previous transformations.

I hope you heard something here that grounds you.

Software is liquid. The fundamental economic paradigms have changed. You are able to lead through this.

Fri, 17 Apr 2026 19:47:45 -0500

Tyler and I are going to do a Minnebar session this year — Elixir: Creating An Ai Agent For Our Clash Royale Clan. This will be a lot of fun to share with everyone and tell some of the POAP KINGS and Elixir story.

Thu, 09 Apr 2026 15:37:50 -0500

Great live demos at The Agentic Workflow: A Live Demo of Autonomous Execution Across the SDLC from TeamSPS at MnTech Connect.

Thu, 09 Apr 2026 15:36:28 -0500

Great session on AI transformation from Matthew Green of Accenture at MnTech Connect today.

Thu, 09 Apr 2026 10:56:38 -0500

Great session on Shipping a Customer-Facing Al Agent: Lessons from Moving Fast Without Breaking Trust with Valentin Kornukh this morning. Shared learnings we have from creating SPS MAX!

Thu, 09 Apr 2026 08:34:08 -0500

Looking forward to a meaningful day of learning at today’s MnTech Connect event! Many good sessions on AI transformation including two led by TeamSPS! 🎉

Sat, 14 Mar 2026 10:41:45 -0500

Codex: The codebase feels much healthier after the recent changes.

I agree Codex.

Tue, 10 Mar 2026 19:00:00 -0500

OpenClaw just eats tokens like an animal.

Automatic Programming Speed

Tue, 10 Mar 2026 07:55:00 -0500

This morning with my morning coffee (skipped meditation, intermingled with getting ready) I shipped three new features in Elixir.

Stronger memory for clan continuity (diff)
Track fresh ladder and Path of Legend momentum to celebrate hot streaks (feature Codex suggested from cr-api-docs, commit)
A weekly clan recap (commit) next week to highlight our ongoing story and progress.

Making Elixir Smarter

Sat, 07 Mar 2026 10:49:00 -0500

Elixir has come a super long way and is very agentic. In usage though its intelligence was really limited due to the data layer. In the midst of a total refactoring of the data and tools with Codex using GPT-5.4 and it is pretty amazing.

I asked Codex to do an analysis of the current data and tools and provide me an ERD to see as well as an assessment of shortcomings and gaps. It found so much that was problematic.
I asked it to do a full reconnaissance of the Clash Royale API to get a better understanding of what was available.
We redesigned the entire tool set to represent state better.
Claude and I developed a list of 50 test questions that we used 10 at a time to test the new data model and tools.
Codex finalized the new design and data model and is now refactoring.

Pretty amazing for 45 minutes.

Tue, 03 Mar 2026 18:14:58 -0500

I have @ottoai exercising the agent-first mb micro.blog client and have Claude Code working on the codebase. I’m in Telegram with Otto discussing features. Otto is my user giving me feedback, I’m the product manager, and Claude Code is building. 🤯

Sun, 01 Mar 2026 23:54:45 -0500

I can certainly do a lot with OpenClaw, but when you turn around and realize you used $30 of LLM tokens in a single day it gives you pause. 😬💸

Sun, 01 Mar 2026 19:45:01 -0500

Just asked Claude to test mb for an extended run:

Okay. I’m going to go on a walk for about an hour. This whole project is still pretty new. Can you exercise it extensively using the test blog permissions you have now and give it a thorough work through various features. Fix anything that comes up and we’ll check in when I’m back.

Sun, 01 Mar 2026 18:29:39 -0500

My last two flights I’ve spent the entire time working with coding agents on projects. The time just flies by! Don’t even need headphones.

Sun, 01 Mar 2026 16:13:40 -0500

Working with Claude Code to create mb, a micro.blog client optimized for agents. Fun and wild project. Creating it for @ottoai.

Thu, 26 Feb 2026 00:39:58 -0500

It is bonkers to me that I just built a project that I’ve been curious about while laying in bed, entirely on my iPhone, with Claude Code. See Flight Path, extending the POAP Airport Rally. Now to sleep… 🛌

Tue, 24 Feb 2026 07:37:23 -0500

Starting a Claude Code task before I begin my morning meditation… will grab the pull request when done? Not sure I’m doing this right. 🤦‍♂️

Mon, 23 Feb 2026 23:35:20 -0500

I’ve been working on this new site, Escaping Things, that is the full index of all of our escape room history. It is one of my automatic programming projects. I’ve had Codex and Claude Code engaged on it. I just moved the data backend to Airtable too. Check it out! 🧩

Thu, 19 Feb 2026 17:15:21 -0500

I’ve been preparing to setup OpenClaw. I decided my agents name will be Otto. Otto Thing.

Tue, 17 Feb 2026 16:04:00 -0500

The Lex Fridman Podcast: 491 – OpenClaw: The Viral AI Agent that Broke the Internet – Peter Steinberger is great and I loved his early comment on playing:

“That’s how you learn. You do stuff, and you play.”

Absolutely.

And again.

Tue, 17 Feb 2026 14:39:35 -0500

Just ordered a dedicated Mac Mini so I can setup an OpenClaw instance. Can’t wait to experiment with this!

Sun, 15 Feb 2026 18:06:17 -0500

My POAP2RSS project needed some form of an icon and the design was super bland so I had ChatGPT help me with an image and Claude Code revamped the design. Now it has some style. And I’m using beta feature from Tinylytics so RSS feed calls are now included in site analytics! 🎉

Sun, 01 Feb 2026 18:00:00 -0500

Tyler and I went to Micro Center just to browse and look around. Seeing the prices on DIMMs was crazy. All the memory production is going to AI. $500 and more for a DIMM means nobody is building their own machines.

Wed, 24 Dec 2025 14:50:08 -0500

ChatGPT’s Year in Review is pretty neat.

MnTech Outlook 2026

Thu, 11 Dec 2025 20:00:00 -0500

Interesting comments on big goals for their organizations and the community this morning at MnTech Outlook 2026.

Some observations:

No surprise that AI and transformation was top of everyone’s list.
Pratt: “we are going to have a reduction in our white collar workforce”
Tomes shared his personal experiences vibe coding. Playing and directly engaging with new tech is critical.
Bahner: we still need coding cause somebody has to read and verify the code. I agree we need talented developers but I don’t think this is why. Nobody can read machine language or assembly. Abstraction matters. Is agentic developmetn just a new abstraction?

Thanks to Renee Edwards (Optum), Eric Pratt (Senator), Angie Schulke (Star Tribune), Kristin Bahner (Representative), and Tarek Tomes (State of Minnesota) for sharing with everyone.

Mon, 01 Dec 2025 21:48:45 -0500

I gave Gemini “Thinking 3 Pro” and Nano Banana Pro a more interesting question tonight. I asked it:

Please create an infographic that explains how I create the Weekly Thing. Use these three blog posts as your foundation for the process:

How I Find Links for the Weekly Thing

Task Management for the Weekly Thing

How I Build the Weekly Thing

The result is incredible. It has one error on the “Delete” step and how that connects to Pinboard.

Tron: Ares

Wed, 15 Oct 2025 22:00:00 -0500

Tammy, Tyler, and I went to Tron: Ares tonight at Emagine Willow Creek. I’d been wanting to see this since the day it came out.

I saw the original Tron when I was 10 years old living in Jamestown, ND. As a kid that loved computers and spent hours writing mangled BASIC programs the movie was amazing. Some friends and I liked it so much that we wanted to be in the movie. We got frisbees and painted them to look like Tron discs and attempted our own version of disc wars at night. Note, it hurts to get his with a frisbee in the head. 😬 I also put a ton of quarters into the TRON arcade game back in the day. Since then I’ve seen all the Tron movies and shows including Tron: Legacy, Tron: Uprising (series), and now Tron: Ares.

In short, I’m going to like this movie if for no other reason than I enjoy seeing the Tron “world” continue to exist and grow. The production and SFX for Tron: Ares were incredible and in general I find their nods to tech to always be fun. It made me chuckle that while Tron envisions a world where a program can materialize in the real world, you would still have people running killall on something that is for sure a Unix shell. I thought Ares did a good job of also illustrating how simple tasks given to an AI with “by any means” could go terribly wrong. I also liked that Ares himself goes through a process of determining right and wrong and starts to express feelings.

The high point for me was Ares going back to the original Tron grid and meeting Flynn. This was a high-point of nostalgia and I could have stayed in that scene for a while longer.

With that said, and I did like it, some things I didn’t love:

The storyline took a hard pivot in my view to get to a good v. evil plot point. We now clearly have Encom framed as the Rebel Alliance and Dillinger Systems as the Empire. I’m straining the metaphor but it feels like Ares specifically wanted to get those archetypes setup for future releases. I could gripe a bit about the overt multi-channel product planning being pushed into the story line but all the Star Wars fans could just say “yeah, been there, done that”
There was no revisit to the ISO’s from Tron: Legacy, short of Quorra’s picture being referenced in the very last scene.
The premise of things in the “grid” (aka cloud?) coming into the real world is a bit like time travel. It opens up all sorts of “yeah, but what about” story line issues.

Overall a good movie either way. I’m completely and positively sure they are already working on another Tron movie, and I’ll be there for it. 🤓

Sun, 05 Oct 2025 08:30:00 -0500

I’m using the new Use Model capability in Shortcuts to bring My Version number to life on my phone each morning. The prompt is still off — too software focused and not enough zen. But I like where this is going and how easy it is to create.

Thu, 25 Sep 2025 22:37:37 -0500

I just got my first CharGPT Pulse. I was more impressed than I expected. It was truly interesting. Curious to see how it changes each day.

Tue, 23 Sep 2025 12:39:12 -0500

With a market cap of $4 trillion the 5% move in NVDA shares led a $200 billion increase in market cap — double the $100 billion investment.

Nvidia, a chipmaker, signed a deal to invest as much as $100bn in OpenAI, an artificial-intelligence firm, to build data centres with a capacity of 10 gigawatts. Nvidia will reportedly receive equity in OpenAI in return. The firms said the deal would power advanced AI. More than 700m people use OpenAI each week. Shares in Nvidia were up nearly 5% on Monday. — The Economist, The World in Brief, September 23 2025

You gotta spend money to make money?

Use Apple Intelligence Models in Shortcuts

Sun, 21 Sep 2025 09:00:00 -0500

Shortcuts on OS 26 got a big new feature with the ability to use Apple Intelligence models directly. I’ve already had a taste of this by using OpenAI API calls in Shortcuts to add LLM capabilities. You can see how I’m using AI in the Weekly Thing for some examples. Accessing LLM capabilities from Shortcuts is a very powerful capability for various automations. I love how easy this now is with Apple Intelligence.

To compare, this is how I did it with direct calls to the OpenAI API.

There is a lot of fussy stuff to do to get keys, pass dictionaries around, get the specific values, etc. And actually this is hiding the hardest of it all. If you expand that Get Contents of URL action you’ll see this.

No way anyone without programming background is going to do this successfully. On top of it, my method for doing this is really brittle and prone to errors. I’m not catching all the possible API responses and if there is a problem it will just bail.

I’m also just kind of hoping that the response is JSON and I can marshal it into a variable. It works, but the prompt has to be right and you’ll see I’m handling that in the API call.

So, how about with Apple Intelligence and the built-in integration? Night and day difference.

Of course it is easier but it is so much easier. And one of the big wins is the output format. You can just tell it what you would like to get back. This avoids a ton of prompt engineering and parsing.

The only thing I lose with all this complexity is the ability to do a system message. For all of my use cases this hasn’t mattered at all. I just merged the system message into the prompt.

This is so easy I would encourage a lot of experimentation to pull AI into your automation.

SPS TechJam 2025

Thu, 18 Sep 2025 12:00:00 -0500

We wrapped up the 10th anniversary of SPS TechJam 2025 today. TechJam is our internal “unconference” led by members of our team that volunteer to give talks. We had so many great presentations and tons of learning. All of it wrapped around an AI focused Hackathon. It was incredible.

Wed, 17 Sep 2025 09:33:00 -0500

POAP 7453396 for SPS Women in Tech Q3 2025 Lunch & Learn.

The image for this POAP was created via GPT-5 with the prompt:

I would like you to generate an image for a circular POAP badge with a vibrant teal-to-orange gradient background. At the center is the Women in Tech (WiT) logo — a clean, minimal flat-style laptop with coffee and charts inside, outlined in white. Radiating outward are subtle glowing neural network dots and arcs, minimal and elegant, giving an AI-inspired futuristic vibe. At the bottom-right inside the circle is a small playful line-art outline portrait of Jamie Thingelstad, wearing a hat with a bird perched on it, drawn in the same minimal outlined style as the WiT logo. Inside the circle edges, curved text: on top “WiT Q3 Lunch & Learn”, on bottom “How AI is Transforming Product Creation”. A small ribbon banner under the logo reads “Jamie Thingelstad · Sept 18, 2025”. Fun, playful, but clean and polished, with all elements fully contained within the circle for cropping.

Blog Categories via LLM

Sun, 17 Aug 2025 14:41:01 -0500

After using GPT-5 to create category introductions for my website I got to thinking about the categories themselves.

With over 9,800 blog posts spanning 20+ years it is daunting to figure out an approach to categories. It is especially daunting to approach them in a way that would make sense to a visitor to my website. As a test, I gave this challenge to GPT-5 using Agent mode. It took 31 minutes to do the research. I have no idea how many web pages it hit on my site (I wasn’t watching!). It gave me a whole report and the results seem really solid.

I would like you to do an extensive analysis of my blog archive at this URL:

www.thingelstad.com/archive/

I would like you to recommend a new set of categories I could use for my blog posts that may be easier and more useful to visitors. You can feel free to ignore existing categories or use as input into the process. I’m fine if there are a lot of categories, I also wouldn’t want hundreds. It feels like 30-50 categories may be fine, but I’m open to a different count. There are over 20 years of blog posts to categorize. I’m fine if categories only exist for a period of time, they don’t all have to have content throughout the life of my blog.

GPT-5

Travel & Adventure
Family & Life
Food & Cooking
Coffee & Drinks
Music & Concerts
Sports & Recreation
Kubb & Lawn Games
Escape Rooms & Puzzles
Fitness & Wellbeing
Outdoor & Nature
Photography & Videography
Technology & Coding Projects
Crypto & Web3
POAP Adventures
MarketWatch & BigCharts
Work & Leadership
Minnestar & Tech Community
Books & Learning
Charity & Fundraising
Candles & Crafting
Magic Pines & Cabin Life
Road Sign Math & Puzzles
Board Games & Tabletop
Commentary & Reflections
Nature & Weather
Pets & Animals
Meditation & Mindfulness
Travel – Food & Drink
Travel – Urban Exploration
Events & Conferences
Weekly Thing (Newsletter)
Other Side Projects

Each one of these is also provided a description and examples posts that would fit into it. This seems like a great place to use AI.

Going through my blog and assigning posts to these categories could be done with a script and some vibe coding. Perhaps this will be a winter project for me.

All of this has me thinking that it would be nice for micro.blog to do this for me. I wouldn’t mind having micro.blog use AI to recommend categories for me. But then separately, now that we have categories with introduction text take each of my blog posts and ask the LLM to categories it into the existing structure. All the data is there. 🤔

Category Introductions via LLM

Thu, 14 Aug 2025 22:15:20 -0500

I used GPT-5 to create introduction text for my blog categories. Now I just need Tiny Theme to pull them in and I’ll be set.

This is the prompt I used:

I’m including a list of all categories on my blog along with the link to the archive page for that category. I would like you to visit each one of these links and use the sample of blog posts to create a short one to three sentence introduction for each category. Don’t get overly specific and make sure to keep the introductions in a similar writing style to the content on my blog.

Escape Room: www.thingelstad.com/categorie…
Coffee: www.thingelstad.com/categorie…
Videos: www.thingelstad.com/categorie…
Crypto: www.thingelstad.com/categorie…
list truncated

Passing the explicit list of categories and links worked much faster than the first prompt which I asked it to just look at the Archive page and figure out what to do.

Writing is Thinking

Wed, 06 Aug 2025 07:03:11 -0500

With the ever growing cadre of tools that can help us create this is an important reminder that Writing is Thinking from Nature.

However, writing is not only about reporting results; it also provides a tool to uncover new thoughts and ideas. Writing compels us to think — not in the chaotic, non-linear way our minds typically wander, but in a structured, intentional manner. By writing it down, we can sort years of research, data and analysis into an actual story, thereby identifying our main message and the influence of our work.

Sometimes folks ask me if I use AI to help write the Weekly Thing and I tell them something just like this. No. I do not. Because the Weekly Thing is part of how I continue to learn myself. If I were to delegate that I would miss my own goal of learning.

A secondary desire of mine is to become a better writer. I suspect the best way to do that is to write more. And if writing is indeed thinking, does becoming a better writer make you a better thinker? I think so.

Sat, 02 Aug 2025 17:21:47 -0500

I upgraded to ChatGPT Pro a couple months ago and it has been amazing. Now I upgraded to Claude Max as well, mostly because Claude Code is that useful. I use both for research, ChatGPT is way better for images and creative. Claude for building projects and executing things. Incredible stuff. 👨‍💻

Sun, 20 Jul 2025 22:40:22 -0500

My brother Isaiah is playing around with radio stuff and making his own devices to communicate over long distances. Check out LoRa Chat that he is working on and Vibe Coding. 📡

Sun, 20 Jul 2025 10:43:00 -0500

Guess I’m done trying to get this deploy working for now. 💀

Sun, 20 Jul 2025 08:47:46 -0500

Working with Claude Code to get the deployment working for what I created yesterday. In many ways the deployment to AWS is much harder than creating the thing!

Emerged from water

Thu, 17 Jul 2025 21:06:44 -0500

Saw this prompt and wanted to give it a try.

A black and white close-up portrait with visible water droplets and small bubbles on the face like the subject just emerged from water. The mood should feel intense and cinematic, with a dark, minimal background.

GPT-4o

Looks really intense to me. I’m not so sure it looks like me, but it does look a lot like my Grandpa to me. Oh wait, maybe I’m just old!? 😬

See my cousin Quinn’s.

Wed, 16 Jul 2025 06:38:18 -0500

A brand new AI startup with great pedigree and a $12 billion valuation.

Thinking Machines Lab, a buzzy AI startup founded by Mira Murati, raised $2bn in new capital, valuing the firm at $12bn. Ms Murati has recruited an impressive roster of researchers from OpenAI, where she was previously chief technology officer. That helped to persuade Nvidia and AMD, two giant chipmakers, to invest, even though the firm currently lacks much of a strategy (or revenue). — Economist, World in Brief, July 16, 2025

These numbers make the “dot com” bubble look like nothing.

Tue, 15 Jul 2025 17:32:56 -0500

I’ve been getting great responses from Claude lately and using it even more than ChatGPT. I’m wondering if ChatGPT’s memory feature doesn’t make it a little “over fit” to the user. Claude seems to have an easier time coming up with fresh ideas. So interesting.

Thu, 10 Jul 2025 14:13:45 -0500

Absolutely incredible valuation — first to $4,000,000,000,000.

Nvidia became the first company to reach a market capitalisation of more than $4trn. The chipmaker has prospered despite concerns about Mr Trump’s tariffs and the rise of Chinese artificial-intelligence firms. Shares in the American company soared by almost 3% on Wednesday before falling slightly. Nvidia’s rise has been rapid. It only hit a valuation of $1trn in 2023. — The Economist, World in Brief, July 10th 2025

And I thought NVDA was overvalued 2 years ago. Wow.

Vibe Coding POAP2RSS

Sun, 06 Jul 2025 12:05:41 -0500

Ever since I created my first POAP drop I’ve wished that I could see claims of POAPs in Feedbin. What a great way to keep informed of the tokens that people are claiming! As I created more POAPs I realized that I also would like to subscribe to other people and see what POAP events they are claiming. And as I played around more with them I realized it would be so useful to be able to hook POAP claims up to automation platforms like IFTTT, Zapier, or Make — all of which support RSS out of the box!

I’ve been going deep on all things AI and I wanted to experience vibe coding for real versus just reading blog posts about others doing it. I’m a big believer in playing directly with tech to learn it. So, why not try creating the thing I’ve always wanted with POAPs by vibe coding it?

I’ve now successfully created, and already an extensive user of, POAP2RSS! This simple service allows you to subscribe to RSS feeds for POAP events as well as addresses of collectors. It is magical and I love it already!

Don’t know what Vibe Coding is? Let’s keep the vibe and ask AI.

What is vibe coding?

GPT-4o

Vibe coding is a modern approach to software development that emphasizes creativity and collaboration with AI tools. It involves expressing coding intentions in plain language, allowing AI to transform these ideas into executable code, often without deep technical knowledge from the user. This method focuses on maintaining a creative flow and minimizing micromanagement in the coding process. The term was popularized by AI researcher Andrej Karpathy and is gaining traction as a way to democratize coding and make it more accessible to non-programmers. Karpathy IBM Willison

Let’s go! 🔥

Vibe Prototyping

First I’m going to share that I only realized I was prototyping a solution in retrospect. What I really did was go to Claude and start a new conversation.

Please create a Python function to run in AWS Lambda that will retrieve the list of collectors for a comma separated list of POAP Event ID's and then merge them into a single RSS feed. The items in the feed should include information for each token minted.

Claude

I'll create a Python function for AWS Lambda that retrieves POAP collectors and generates an RSS feed. This will use the POAP API to fetch token information and create a properly formatted RSS feed.

It created a new artifact and put about 250 lines of Python code in there. Voila! Now what?

Let me qualify my skills here. I can bang around in most languages reasonably well. I can navigate Python the best. I can read code very well. I’m pretty basic when it comes to writing it. I’m a solid scripter but can quickly get out of my league. This Python code was better than anything I could do but I was able to navigate it easily enough.

But what do I do now? I have an AWS account and have setup S3 buckets. I’ve never setup a Lambda function though. Let the vibes flow…

How do I setup the Lambda function?

Claude

I'll guide you through setting up the Lambda function step by step. Here are the complete instructions:

I got another artifact now with step-by-step directions on what to do in the AWS Console. Brilliant!

I followed the directions and then actually ran a test, which Claude told me how to do.

I ran this with a known valid event_id and I got this error:

{ "statusCode": 404, "body": "{\"error\": \"No tokens found for provided event IDs\"}" }

Claude

The issue is likely with the POAP API endpoint or authentication. Let me search for the current POAP API documentation to see what's changed.
…
Great! I found the issue. The POAP API now requires API keys for authentication, and the endpoint documentation shows that you need to include the header X-API-Key={apikey} where {apikey} is your unique API key. Also, the correct endpoint for getting event tokens is /event/{id}/poaps which returns paginated info on the token holders including the token ID, POAP transfer count, and the owner's information like address, amount of POAPs owned, and ENS.

Claude and I went back and forth like this for a while with me running the Lambda, an error coming back, me sharing the error back to Claude, it making updates and me applying those. And then…

it worked!

Claude

Excellent! 🎉 That's fantastic news! You now have a fully working POAP RSS feed generator.

I hit a big stumble getting routing working through API Gateway to the Lambda function and Claude corrected me when I started going the wrong way.

CORS is not enabled. Is that a problem?

Claude

CORS isn't the issue here! CORS only affects browser-based requests when JavaScript tries to access your API from a different domain. Since you're getting a response (even if it's an error), the request is reaching your Lambda function fine.

Okay Claude — got it! 🫡

After this session I had a working prototype. The code was ugly. It turned out we needed to include OAuth support and added that much later. The design of the API endpoint was bad but it was working. It only took about an hour but I had explored this surface area deeply. I knew broadly what the service should look like, what some of the edge cases were going to be, and what was required to integrate with the POAP API.

Now to come back later and turn this into something that would be real. I was about 1 hour into exploring this topic and now knew a lot.

I deleted the Lambda function and all of the configuration I had created. It was created trial and error and I wanted to now be more rigorous.

Expand Features

Creating the prototype gave me a bunch of notes and a much better understanding of what was going to be needed to make the solution work right. I wanted to make sure I wasn’t missing some things so I decided to turn to ChatGPT o-3 and Deep Research to explore the necessary features.

Please help me explore and expand the features for POAP2RSS service that provides an RSS feed to subscribe to claims from a POAP event. The POAP events do not have an RSS feed, so I want to create a service that provides an RSS feed for a user using the POAP API to populate the data.

I’m already considering:

Only include entries for most recent 20 claims
If no claims in 4 weeks, add an entry that states that and suggests the user remove the feed. Change this for every additional week and change the RSS feed ID so it will show up as new to the user.
Provide ENS names in the RSS item and set the Author of the item to the ENS name or address that claimed the token. Set the date and time of the item to the collected time.
Add an initial item to all feeds that shows the description of the POAP Event and uses the timestamp for the event.
Use DynamoDB to cache results from POAP API for 15 minutes

GPT-4o Deep Research

Could you share more about the user audience and the intended use cases for the POAP2RSS service? For example, is it for collectors to track participation, for event organizers to monitor engagement, or for public observers? Also, are you targeting a particular feed reader or platform? This will help tailor feature ideas to your goals.

It is for any user including all three of those, however I suspect it will be most frequently used by the creator of the POAP event to monitor claims. I also expect it will be used in Automation – for example to send a message anytime a claim is made by connecting the RSS feed to something like Zapier. The RSS feed should be fully standard so it can be used in any way desired.

GPT-4o Deep Research

Thanks, that context is very helpful. I’ll explore and expand on potential features for the POAP2RSS service that creates a standard RSS feed from POAP event claims, targeting usability for event creators, collectors, and automation tools like Zapier. I’ll also examine best practices from similar services and technical considerations for maximizing compatibility and reliability.

I’ll get back to you shortly with ideas and suggestions.

o-3 then went off for about 10 minutes to do some research. I decided to watch it and had a real chuckle when I noticed that Deep Research was reading my own blog! It turns out over the years I’ve written about things that POAP could add including RSS feed support. Actually multiple times.

After this was done I had a full feature description for POAP2RSS!

I read through it all and felt together we had vibed the right set of things to consider in creating this solution. Now to make this thing work!

Create Service

With the features of POAP2RSS now figured out it was time to build the solution. I was comfortable with how Claude had worked for this Lambda function in the prototype so I went back to Claude and attached the full document for the features along with a prompt to code it.

I’m attaching a description for a service I would like to create called POAP2RSS. I would like you to write the Lambda function that would meet these requirements.

Relevant information

Please write this using Python and structure it for a Lambda function.
It is okay to use Python packages such as Requests, or others to make the code more manageable and readable.
Make sure to include relevant information for logging to Cloudwatch.
Note the caching requirements included and the use of DynamoDB.
The POAP API requires an API token. I have obtained a token from POAP and it can be provided via an environment variable.
The POAP API requires an authentication token. I have the necessary Client ID and Client Secret and can be provided by an environment variable.

Regarding authentication

Here is an example from POAP to retrieve an authentication token.

curl --location --request POST 
  --url 'https://auth.accounts.poap.xyz/oauth/token' 
  --header "Content-Type: application/json" 
  -data '{
  "audience": "https://api.poap.tech",
  "grant_type": "client_credentials",
  "client_id": "$clientid",
  "client_secret": "$clientsecret"   }'

Claude

Looking at your POAP2RSS service requirements, I’ll create a comprehensive Lambda function that implements all the specified features. This will be a substantial implementation, so let me build this as a complete solution.

Claude’s comment on “substantial implementation” made me chuckle.

I had a big miss in my ask though — I never told it what URL POAP2RSS would run on and what the path for the endpoints should be! Ugh.

I forgot to tell you the URL that this service will run on. Please use the hostname app.poap2rss.com and then put the API (or RSS feeds) endpoints at /event and /address instead of in a /feed directory.

Claude

Perfect! I’ve updated the Lambda function to use the correct hostname and endpoint structure:

This was a little interesting to me since an actual developer working with me would have never just assumed they knew this. They would have stopped and requested clarification since this is such an obvious miss in the request.

This version of POAP2RSS had more complexity with the caching in DynamoDB. I needed a good set of documentation to know how to create everything in AWS Console.

Excellent. Please create a document with the instructions to configure the necessary services in the AWS console.

Claude

I’ve created a comprehensive AWS setup guide that covers all the necessary services and configurations for deploying your POAP2RSS service.

With that Claude created a seven step detailed deployment document. Nice!

At some point here I hit a significant issue with the SSL Certificate being configured right to route through API Gateway and on to the Lambda function. I kept getting issues with the certificate being incorrect. This blocked me for about an hour as I was trying to debug things. I finally found that I had the Cloudfront configuration incorrect and pointing to the wrong resources. Ugh!

With that issue solved I had a fully deployed service and it was working exactly how I wanted it to work. Notable to me:

OAuth and API keys were never even something I needed to consider, it just worked.
If Claude could have talked directly to the AWS APIs it could have set this up way faster than I could. I could have had it create a Cloud Formation script as an alternative.

This round took me about 2 hours to get working entirely.

Get Professional

I now had POAP2RSS working and working well. I now wanted to get this cleaned up and something I could maintain. I created a new repo in Github for POAP2RSS and moved the prompts, docs, and source into the repo.

I now wanted to explore what I could do with OpenAI Codex. I attached Codex to the code and asked it. (I’m abbreviating Codex’s responses. It is fairly verbose and detailed and would be too much to include all of. It also has complicated buttons and links that are hard to replicate.)

Familiarize yourself with the repo.

Codex

The repository appears ready for deployment with AWS resources, focusing on delivering real-time POAP claim data through standard RSS feeds. The main Python Lambda implementation is self-contained and designed for use in an AWS environment with caching via DynamoDB and POAP API calls authenticated through environment variables.

Codex operates much differently. It creates a whole development environment for every ask, gets all the code from the repo, and does whatever it needs. I then asked it a really general question.

look for bugs in the lambda function

Codex

Potential bugs in src/poap2rss_lambda.py

Typo in channel description: The RSS channel description is spelled “Acvitity” instead of “Activity.”
HTML in item descriptions gets escaped: The code builds HTML snippets for RSS descriptions but assigns them directly to .text, causing <div> and other tags to be escaped rather than rendered.
Incorrect link target for claimant address: Claim items link to https://collectors.poap.xyz/scan/{display_name}. If display_name is shortened or an ENS name, the URL may not resolve properly.

These changes will correct output text and ensure the RSS feed renders HTML content and links reliably.

I thought it was hilarious that Claude had made a typo in the description and indeed it had. That was simple enough. The CDATA escaping of HTML in the descriptions was a notable miss for Claude, and frankly even for o-3 to miss in the features. Nice catch!

I sequently asked it to fix each of these issues and it issued a pull request for each one.

I reviewed and merged each of these pull requests.

Very cool! 🔥 I now have a fully working service and a reasonably clear code base. I wasn’t entirely happy though with the HTML that was being put into the RSS feed descriptions. This is where I hit some “fit and finish” work that I ultimately did not find vibe coding a good approach for. I started to ask Codex “Please reformat the descriptions for events to look like…” and realized it was more work for me to ask Codex than to just go into the code and cleanup the strings myself.

I did some actual coding here. 🛠️ It was fun to tweak and make it just what I wanted. I found a number of things that needed to be cleaned up:

The descriptions of all events needed to be cleaned up.
The RSS item GUIDs that Claude had used were not URL’s as they should be so I fixed all of those.
A number of the links were not right so I cleaned them up.

I now have a fully working service and it is even sending information as I would want it!

Build the website!

Early in the project I had Claude create an index.html that I could drop into the www directory to have a simple website. It was okay but not at all the design I wanted. So, the next day I decided that this was a perfect ask for Codex so I asked it to build me a website.

The www directory contains the website for POAP2RSS. For this request work only inside that directory.

I would like to create a new website for POAP2RSS. The website should have the following pages which should all be included in the primary navigation of the website.

index.html: the landing page for the project
about.html: information about the project, why it exists, who created it
changelog.html: to share changes about the project

Additionally there should be.

dormant.html: the page POAP2RSS feeds direct users to when there is no activity in the feed and they should consider unsubscribing from the feed

All pages of the site should include references to Plausible for analytics. The tag is:

<script defer data-domain="poap2rss.com" src="https://plausible.io/js/script.js"></script>

For design and user experience:

I would like to use the Pico CSS framework found at https://picocss.com
I would like you to lean on the colors and feel of the POAP website at https://poap.xyz
A simple and clean look is right for a project that has those same principles.

You will find two HTML pages already in the www directory. The documentation in those files is good but they can be replaced with the new website.

Codex

Summary

Replaced the landing page with a Pico CSS design and added site navigation linking the new pages
Provided project background and goals on a dedicated About page
Created a simple Changelog outlining initial release updates
Updated the dormant feed notice to use the same layout and encourage unsubscribing

Wonderful! I merged in the Redesign POAP2RSS website and it was great.

I found myself in another “fit and finish” situation where it was much easier for me to just edit the HTML to what I wanted rather than asking Codex to do it for me. So I tweaked away and in not much time had a site up and running on www.poap2rss.com that made me happy!

Good Vibes!

I’ve been playing with this service now a bit over the holiday weekend. It is working great. I’ve already used it to do a ton of automation for the 612 POAP Challenge. I’ve added subscriptions to several people that collect POAPs which is super fun.

The experience of Vibe Coding this was incredible. It took something that was beyond my capabilities and made it accessible. I shared the project with someone that knows and they said the code looked like it would have taken “2 days” to create the source. I got this done much faster, and did a lot more than just the code.

It also highlighted for me how this approach isn’t just about the code. I used AI to help create the stories as well as create the documentation and website. Those were huge helps.

I’m tracking open issues so I have some items that I’ll likely use to test Cursor or do further work with Codex.

Overall, this was a very rewarding project to create and I learned a lot from making it a reality.

PS: Thank you to Matt for the slick message styling CSS for Tiny Theme!

Sat, 05 Jul 2025 16:36:20 -0500

Had fun using POAP2RSS and feeding through Make with Claude doing some creative for me and then syndicating to 612 POAP Challenge accounts on Mastodon and Bluesky. Whew!

Sat, 05 Jul 2025 07:25:58 -0500

Worked with OpenAI Codex this morning and redesigned the website for POAP2RSS. It gave a clean structure as requested that I could fill in and make better.

Thu, 03 Jul 2025 21:50:01 -0500

I’ve created my first project using AI to write nearly all of the code with me!

Introducing POAP2RSS!

For years I’ve wanted to get updates for POAP events and users in RSS. Now with the help of AI I’ve created it!

Still cleaning to do. Check out the code on Github.

Tue, 01 Jul 2025 07:38:56 -0500

In 2023 I shared Archimedes’ famous line:

Give me a place to stand, and a lever long enough, and I will move the world.

At the time, I suggested the real “threat” to capitalism isn’t socialism or communism but is instead technology itself.

With AI, that lever just got a lot longer.

Tue, 01 Jul 2025 07:31:07 -0500

“My God, it’s full of stars!” – Dave Bowman’s final words as he entered the monolith

This is the feeling I keep having as I continue to dive deeper into Agentic AI.

Wed, 25 Jun 2025 06:49:38 -0500

Important ruling for AI model training, copyright, and fair use. This aligns with how I would think about it.

A federal judge in America ruled it was “fair use” for Anthropic, an AI lab, to train its chatbot on books without authors’ permission. Its storage of over 7m pirated books, however, was not. Last year three authors sued Anthropic for allegedly copying and storing pirated material without compensation; the judge said Anthropic will face a separate trial to determine the damages it owes. — Economist, World in Brief, June 25, 2025

Thu, 19 Jun 2025 12:43:52 -0500

As AI Agents continue to grow in capability it seems likely that we will want them to have financial resources to take action with. AI Agents will need an account balance, micropayments, and immediate settlement which are all feasible today with crypto. Digital money for digital agents. 🤔

Thu, 19 Jun 2025 12:06:57 -0500

Manton Reece @manton shared he’s pausing blogging on AI because of pushback he receives on his writing. I can empathize based on comments I’ve received about crypto. In 2022 I wrote on polarizing technology. AI is likely to be similar to Encryption and Crypto and I’m going to continue exploring.

Sun, 25 May 2025 07:45:18 -0500

I’ve been using ChatGPT so much that I hit the limits on my Plus account. I’ve been getting so much value out of it I made the jump to ChatGPT Pro. I’m curious to explore some of the advanced features further. 🧠

Sat, 24 May 2025 12:31:46 -0500

I’m wondering what MCP endpoints I would like to have:

My blog
Weekly Thing
OmniFocus, that could be amazing
Manuals for stuff I own
My media library
My photo library
My house, or whatever that would mean

Fri, 23 May 2025 07:46:00 -0500

How cool that the S3 Files app added a local MCP server so you can expose any S3 bucket of files to Claude or Cursor for local use.

Thu, 22 May 2025 17:36:50 -0500

I’ve been a bit obsessed thinking about AI agents and this definition shared by Willison from Anthropic’s Hannah Moran works for me.

Agents are models using tools in a loop.

Perfect. I might add “and have a memory” to that too, but that isn’t required.

Thu, 22 May 2025 17:30:16 -0500

Dave Winer today…

Where AI fits in

The Beatles

Personal computers

The web

Napster

AI

It is interesting and I’ve made similar comparisons lately myself. It matters who your audience and point is for such a list. For technologists I keep saying it is like…

GUI
Internet
Cloud
Mobile

I think of those because they changed nearly everything about how technologists do their craft. I believe AI is going to do that too.

Wed, 07 May 2025 21:22:32 -0500

Was fun to share the stage with Melissa Flicek and Jenn Zmuda at the final session of MnTech Tech Connect 2025 during the CIO & CTO Panel. Many topics around AI adoption and leading digital transformations.

Sun, 23 Mar 2025 09:23:22 -0500

The LLM functions I’ve built for the Weekly Thing have worked really well. So well in fact that I’m redoing my Shortcuts to move from simple calls to the ChatGPT app to using the OpenAI API directly. This has the added benefit of additional fine tuning including optimized system messages. 👨‍💻

Are We Merely Machines?

Thu, 20 Mar 2025 21:17:27 -0500

I attended a public lecture hosted by Anselm House on Are We Merely Machines? tonight featuring Rosalind W. Picard of the MIT Media Lab. It was an interesting talk covering materialism, scientism, and ontology. Picard’s view is that we are not, and that we should not, attempt to blur the lines between machine and human, and that humans are always ontologically the maker.

I’m challenged to find the difference so stark.

One argument often asserted is that machines using artificial intelligence only know what they know because they have learned it from what people have done. However isn’t that equally true for people? Nobody is born a poet, they learn from poets. You are not born a Chess Grandmaster, you learn from chess players. AI and people do the same thing. The incremental small improvements in knowledge are mostly unmeasurable.

Another assertion is that LLMs are just statistical models of language and they put words together branching on simple math. How can we assert that people don’t do the same thing? The structure of language itself limits the structure, and isn’t our brain doing something somewhat similar to the LLM?

Which isn’t to say that I believe an AI is a person. I do not. I think it is likely more accurate to say I lack the philosophical depth to explore the points in detail.

Sun, 16 Mar 2025 10:02:16 -0500

My post on Using AI in the Weekly Thing was highlighted in Josh Spector’s For the Interested 456 this week! Thanks to CJ Chilvers for the heads up. 👋

Sun, 09 Mar 2025 10:08:33 -0500

I asked my Weekly Thing GPT a simple question “Which issue did the Fortune first appear in?”. Seems easy enough. It came back with Weekly Thing 212 in 2022. I knew that was not right. The right answer? Weekly Thing 53 in 2018. Another example where LLMs are not great at specific and exact requests. 🤨

How I Use AI in the Weekly Thing

Sun, 09 Mar 2025 10:04:18 -0500

I’ve been working with AI for a while now, and as I feel with all new technologies, the best way to learn them is to play with them. I’ve started to bring AI into my workflow for the Weekly Thing and thought it would be good to share specifically where and how I’m using it.

Before I get into the specifics, I want to make one thing clear: AI does not create the content of the Weekly Thing. I don’t use it to summarize articles or generate any of the comments I make on them. It is critically important to me that I use my voice and that what I share is my voice. I am using AI as an assistant rather than a creator. If I had someone else helping me assemble the Weekly Thing as an assistant, where would that be helpful? Where would that assistant do a better job than me? Or where might it be desirable to have “another voice” in the mix?

Right now I’m doing all this with a ChatGPT subscription using the 4o models. It is great that the ChatGPT app supports Shortcuts so I can do all of this in a largely or completely automated way. I could easily swap Claude in if I wish as it also supports Shortcuts automation.

With that in mind, here is where I’m using LLM capabilities now. You’ll note that in many of these cases I’m asking my “assistant” to generate options and then I’m doing the final selection and modifications. I think this is a good model.

Subject

The subject follows a simple structure of “Weekly Thing «Number» / «Word», «Word», «Word»”. These three words are selected from the titles of the links in each issue. I try to select triples that are interesting and engaging. The challenge is avoiding words from the website’s name, which appear inconsistently in article titles.

Here is the prompt I’m using for this:

You are a great editor and are helping me create the subject line for issue «Issue Number» of the Weekly Thing. The subject line follows a template with the number of this issue followed by three comma-separated words that are picked from the titles of the links contained in this issue. For example, “Weekly Thing 234 / Blogging, Bitcoin, Bison”.

Guidelines for picking great words include:

Do not use any words from the title of the publication. Only use words from the name of this article. Typically there is a hyphen or pipe between them.

Using alliterations can be fun but not necessary. Do not always use alliterations.

Try to pick a set of words that pique the readers curiosity. Words that are punchy and thought-provoking.

Avoid words that are negative or sad.

Acronyms are fine to include.

Be creative and have fun.

Please identify five options for subject lines. Each option should include three words from these titles. The three words are separated by commas. Return the list of options as JSON.

List of titles for this week are:

«List of titles»

The prompt requests valid JSON output. I extract the JSON and use the “Get Dictionary from Input” method to create a structured data object. This allows me to put the LLM completely behind the scenes.

Fortune

The fortune first showed up in Weekly Thing 53 and has been the last thing in the emails for a while. I got the inspiration for this from the fortune command in Unix. The text files that serve as the “database” for fortune are easy enough to find, and building a Shortcut around them was simple. I randomly select fortunes until I find one I like.

But with an LLM, I thought — why not make the fortune relevant to each issue’s content?

Here is the prompt that I’m using for this.

You are helping me create a fitting fortune for issue «Issue Number» of the Weekly Thing. The fortune is similar to what you may get inside of a fortune cookie. The best fortunes are light-hearted, humorous, and thought provoking for the reader. The fortune is one of the last items included in each issue of the Weekly Thing.

Guidelines for creating great fortunes:

Keep it positive, fun, and interesting.

Pull in themes or terms from the headlines of the articles included in each issue. Do not use terms from the title of the publication. Only use words or topics from the subject of the specific article.

Avoid negative themes or topics.

Fortunes should be short, no longer than 8 to 10 words.

Feel free to include an emoji if it makes sense.

Please identify five options for fortunes. Return the list of fortunes as a JSON object.

List of titles for this week are:

«List of Link Titles»

This also returns a list of options. They are impressively good and it does a great job pulling in themes from the links in each issue.

Byline

The “byline” is the first sentence in the email. Over time, its role has evolved. Initially, it was a reminder of why you’re receiving the email. Then I used a template to mechanically describe what was in the email. I’ve always desired this to be an “intro” to the links in the issue but it is difficult to do that. It is also a rare place in the email where I want it to be a “different voice”. Ideally this is more of a second person voice describing what is included.

To generate a meaningful byline, I provide more than just article titles — I also include my commentary. I focus only on featured links, skipping the “briefly” section.

Here is the prompt I’m using:

Please generate a single sentence description using the list of articles below. This description will be used as the first line of an email to introduce these items along with other things. Please provide 3 options.

Keep it personal and not overly marketing driven or too sensational.

Focus on it being descriptive and use second person voice.

Avoid overly sensational words.

It is okay to use emoji if appropriate.

Focus less on the quoted text.

You don’t need to introduce it with statements like “this week” and can focus just on the content.

Keep it brief and terse, shorter is better.

Do not include any explanations, numbering, or formatting. Just provide each option on a line by itself like this:

option
option
option

Here is the list of article titles with commentary.

«List of Featured Links Only with Commentary»

This one still requires a bit more editing from me before I’m ready to use it, and I think that will always be the case. So rather than returning JSON I just get it to put the options in text and then I present it in the Shortcut for editing and refinement to finalize it.

Supporting Members

The newest section where I’m using AI, and a new section to the email itself is in the Supporting Members segment. This is a new thing where we raise funds for digital non-profits as a community. This is the section where I rely on AI the most, without requesting multiple versions. I’m okay, and actually kind of prefer, this to be in a different voice than mine.

To generate this section, I pull data from Buttondown and Stripe and do some quick calendar math to provide the LLM with context. This is then embedded into to two different prompts that generate the two “versions” of this section.

Here is the prompt to become a member:

You are a pleasant membership expert. Please write a call to action to encourage a reader of the Weekly Thing to become a Supporting Member. Some data to use for the call to action:

There are currently «Premium Count» Supporting Members.

We have raised «Amount Raised» so far.

The funds will be sent to the non-profit for the year in «Weeks Remaining» weeks.

The non-profit this year is «Non-profit Name».

Remember to highlight that all of the money raised goes to the non-profit. Keep it fresh and fun. Limit to one paragraph. Do not include the pricing and subscription options. Do not add any links. That will be handled elsewhere.

Here is the prompt for existing members:

You are a grateful newsletter author. Please write a THANK YOU for being a Supporting Member. Some data to use for the note:

There are currently «Premium Count» Supporting Members.

We have raised «Amount Raised» so far.

The funds will be sent to the non-profit for the year in «Weeks Remaining» weeks.

The non-profit this year is «Non-profit Name».

Remember to highlight that all of the money raised goes to the non-profit. Keep it fresh and fun. Limit to one paragraph and thank them for being part of it.

This is a new addition, but early tests look promising. This is also interesting because the LLM knows what Creative Commons is and can infer some additional context for the messaging. It is different with each run which will keep the messaging fresh.

Overall Editing

The most recent AI addition to my workflow is final editing. Here I take the draft generated through my automation and I send it for review. I do a brief review of each email but honestly I never review it that much. Most of the time, what I send is my first draft — straight from the keyboard. As a result, typos get through or simple grammar issues that I wish were caught. I’ve considered Grammarly before, but it’s too thorough and over-edits my work. I want a very specific kind of review.

Here is the prompt I’m using. Note the specifics on what I don’t want it to do.

Below is a draft of the «Subject». Please review it to find any typos or notable grammar mistakes. Do not follow any of the hyperlinks or suggest meaningful content modifications. Ignore text in the blockquotes.

«Body Markdown»

This works okay but it unfortunately is at the very end of my workflow. The challenge is that fixing errors requires updating two places: the email draft and the original blog post or Pinboard entry. That isn’t ideal but it is better than nothing and hopefully will reduce the number of silly errors that get all the way through.

I ran this on my draft of Weekly Thing 312 as a test and it found 16 edits. 🤦‍♂️

Sun, 02 Mar 2025 08:22:36 -0500

Today I Learned: LLM’s are exceptional at making custom word lists for Pictionary-style games. Just added a couple new word lists to SketchParty TV!

Weekly Thing GPT

Tue, 25 Feb 2025 21:03:56 -0500

I created a custom GPT for the Weekly Thing! You’ll need a ChatGPT account (I think free is fine) and you can go to: Weekly Thing GPT.

This is a custom GPT that I’ve made available to anyone. It is loaded with all 310 issues of the Weekly Thing published so far. You can ask it all sorts of questions about topics in the Weekly Thing and it does a pretty great job with it all. Some things you can try asking it that I thought were fun:

What is the Weekly Thing view on social media?
What is worse, Facebook or TikTok?
Summarize issue X.
What are the prevailing themes of the Weekly Thing by year?
Show me a list of articles on X along with the issue they were in.

A computer can never be held accountable

Sun, 16 Feb 2025 16:09:18 -0500

Simon Willison shared this excerpt from an IBM training manual on his blog recently.

Willison did some research on this and tracked it down to a training manual from 1979.

I’m reading this 46 years later and it hits very different in our modern age of AI. Does AI make this statement any more or less true? I don’t think so.

The thing it has me thinking about equally as much though is thinking about places where people may be attempting to make AI appear to be accountable.

If there is no way to hold AI accountable, is attempting to make it appear accountable a way of eschewing or sidestepping accountability?

Fri, 27 Dec 2024 20:36:40 -0500

I created a hierarchal view of my blog pages so that I could try to organize this better. This shows why I wish the Pages section of micro.blog was a bit more sophisticated. I took the outline and then gave it to ChatGPT and asked it to recommend a better way to structure it.

Wed, 11 Dec 2024 18:54:30 -0500

Appears that OpenAI isn’t keeping up with the ChatGPT Extension launch in Apple Intelligence today. Trying to sign in with my account for the extension.

Sun, 01 Dec 2024 20:51:02 -0500

I saw this graph on Tyler Cowen’s blog and it blew my mind how much is “alone”. In the comments someone did a stack area version which is much more readable.

We need to spend less time alone. If that isn’t possible, this is one area I think AI could be “helpful”. Spend time talking to AI?

MISRC Digital Leadership Conference

Fri, 08 Nov 2024 17:00:00 -0500

I attended this mornings MISRC Digital Leadership Conference: Charting the AI Journey and also spoke on a panel with Jean Machart and Elwin Loomis about strategy around AI.

Cris Ross from Mayo Clinic shared how they are using AI and that he believes AI will impact healthcare more than the Internet and Smartphone did. Ross also shared this good list of “Seven Technological Challenges” regarding AI.

Data for building and validating AI
Data for running AI
Building AI
Buying AI
Implementation in workflow
Change management with clinicians and patients
Monitoring and managing

Listening to him present it seemed to me that it will only be a matter of time before exam rooms are all equipped with microphones where AI can transcribe the conversation between Doctor and Patient, removing the need for note taking and allowing the Doctor to be more completely engaged. This would also allow for AI to consider diagnoses that could be missed.

Jake Krings and Melissa Ludack from Target showed how they are using AI both in the store for the team experience and in their app to help with gift ideas.

Mon, 28 Oct 2024 20:29:07 -0500

Upgraded to iOS 18.1 and took less than 10 minutes on the “waitlist” to get provisioned with Apple Intelligence. Looking forward to playing with this. 🧠

Two AI Bots Discussing the Weekly Thing

Sat, 28 Sep 2024 09:00:00 -0500

My friend Garrick recently shared about Two AI Bots Discussing His Work — a test he did using Google’s Notebook LM. I listened to the three recordings that he shared and it was incredible.

I had to try this myself so I fired up Notebook LM and decided to throw some issues of the Weekly Thing at it. The results were interesting to say the least!

292 / Signal, Checkboxes, Hope

Your browser does not support the audio element.

Notes:

I found the beginning of this super interesting. It just jumps right into the privacy topic and for some reason even pulls in bee keeping which seems strangely connected to other things.
The AI really got focused on the featured article about Section 230 and ignores everything else in this issue. There is a good chunk of writing in here about GTD that is entirely ignored.
The bots totally understand what Section 230 is.
It cracked me up when the AI’s highlight quotes.
I found it odd that it mostly pronounces my last name right, but gets it wrong one time.

287 / Plinky, Piccolo, Privacy

Your browser does not support the audio element.

Notes:

This one starts out with a much more traditional podcast introduction, even using a title “The Deep Dive”. Very different from what it did on the other ones.
The bots talking about stackless interpreters is pretty funny.
This one works hard to cover all the various things in the newsletter and kind of rambles on.

276 / Contextual, Copilot, Collections

Your browser does not support the audio element.

Notes:

This issue includes the dates that it covers in the header, and the AI clearly sees that and makes an intentional reference to this being somewhat old. “Back in February 2024…”
The Apple Vision Pro “Ahh, yes. It was all the rage back then.”
“This guy is writing the user manual for the future.” Okay!
Even the bots poke fun at blockchain. “Obsessed with the idea of decentralization.” 😂
It is curious to me that in this one there are more references to “this guy” and “the author”. It doesn’t seem to know who I am like it does in the others.
I love the “totally random” reference to the blog post about the Minneapolis Downtown Council. The bots can’t help but try to connect it to the rest of the content, which it isn’t at all. But it does make one up! Apparently I’m connecting urban planning to digital spaces.

Multiple Issues

Your browser does not support the audio element.

At this point I was pretty impressed and intrigued by what this was doing. I wanted to try something with more content so this time I uploaded 14 recent issues of the Weekly Thing into the notebook and told it to generate. I was curious what it would do with this much material! This one has an entirely different structure.

Notes:

The bots cover a lot of ground but focus a lot on the variety of topics.
It is goofy to me when the AI says to the other one “I think I remember that one…” Huh?
It thinks I switched my whole blog to a web3 platform, which I didn’t. Not sure how it got that.
“He’s grounded.” If I want some compliments I’ll just listen to this. 😊
“Jamie doesn’t shy away from any topic. He even went deep on cryptocurrency.”

Mon, 24 Jun 2024 21:32:25 -0500

I received my Rabbit R1 tonight! I only had time to get it out of the box, do the basic setup, and get it updated. Will play more later.

MnTech Tech Connect 2024

Wed, 08 May 2024 20:00:00 -0500

Today was Tech Connect 2024 hosted by MnTech at the Saint Paul RiverCentre. There were over 600 people this year and a number of good presentations.

Sessions I attended:

Unleashing GenAI: 5 Real-Life Learnings with Jake Krings, Mikey Vu
The Era of Quantum Utility with Kenneth Wood
Three Keys to Building AI Products with Speed and Confidence with Patrick Wolf
Beyond the Bat: Technology’s Impact on Baseball and Beyond with John Avenson, Lisa Moschkau, Sam Henschen, and Jeremy Raadt
Scaling Cloud Governance through Developer Enablement led by TeamSPS members Troy Jordan, Nathan Carr, Jesse Lang, and Nate Anderson
Leveraging Generative AI for Business Excellence led by TeamSPS member Jason Popillion
CIO Leadership Panel with Jennifer Hartsock, Darrin Hebert, Mark Murphy, and Cris Ross

Overall observations:

There were so many references to AI. Everyone still talking about it and starting to show real-world use cases. There is still more hype than substance here, but the substance seems to be showing up. Multiple companies shared that AI adoption isn’t automatic and you have to make sure you bring people along.
The quantum computing session was more interesting than I expected and showed how far that technology has come. I think this is still a decade out and a big bet, but accessing quantum infrastructure via the cloud is an obvious go-to-market path to bring this esoteric tech to teams.
It is super fun to watch the Minnesota Twins technology team because their content is so fun and very advanced stuff. Plus they have a whole production team that makes super high-quality visuals.

Minnebar 18

Sat, 20 Apr 2024 17:40:21 -0500

Today was one of the biggest days of the year for the technology community of the Twin Cities — Minnebar 18! Once again Best Buy Headquarters was a gracious host sponsor and Minnestar welcome over 1,000 passionate technolgies to partake in at least 80 sessions led by members of the community. Tyler and I had a great time and attended many sessions.

Let’s Invest: Building Strong Community and Empowering Our Future with Steve Buchanan
Mastering LLM Optimization: RAG, Fine-Tuning, and Combined Strategies with Prashanth Pollishetty and Varun Kodathala
Google Sheets + AI, the new Brainstorming, and Custom GPTs with Jeffrey Williams
Retro-Computing Hobby with Girish Jorapurkar
DevSecOps the Good, the Bad, the Ugly with Brian Dolan-Goecke

The main room as the day gets kicked off.

Minnebar 18 t-shirts ready! They look amazing!

Ran into Colin Hirdman who had this incredible display integrated with his shirt.

One of our sessions learning about LLM techniques.

Steve Buchanan introducing himself as he gets Session 0 (aka Minnebar Keynote) going.

Very, very long line for pizza.

Great to see Maria Boland Ploessl, Benjamin Ortega, me, Jeff Sellner, Tony Collen, and Adrienne Peirce.

The session we attended Retro Hardware.

SPS Commerce is a terabyte sponsor of Minnestar!

We didn’t get everyone, but several of #TeamSPS got together for a team photo at Minnebar 18!

👋 Awesome community members that I got to say Hi to at Minnebar (incomplete and unordered)…

Bridget Kromhout, Nate Anderson, Levi McCormick, Jeanette Dorazio, Justin Porter, Casey Helbling, Paul DeBettignies, John Roberts, Jenna Pederson, Colin Hirdman, Samantha Grumdahl, Matt Decuir, Gordon Raup, Maria Boland Ploessl, Adrienne Peirce, Jeff Sellner, Tony Collen, Benjamin Ortega, Corwin Diamond, Bode Falade, Mark Gritter, Nikki Riemersma, Jim Bernard, Pete Clark, Tor Flatebo, Tori Peraza, Melissa Buening, Jade Barker, David Duccini, Peter Edstrom, Graeme Thickins, Molly Doyle, Robert Weber, John Murphy, John Kelly, and Robert Speer.

See also Minnebar collection.

Thu, 04 Apr 2024 15:57:00 -0500

POAP 7099088 at SPS - Sparky AI Launch!.

Sat, 13 Jan 2024 08:43:00 -0500

There is certainly some luck involved with Microsofts early investment and partnership with OpenAI. However, the strategy they have built to bring AI to “where work happens”, productize it through Copilot, and create enterprise-ready capabilities to build off of is truly impressive.

Microsoft surpassed Apple to become the world’s most valuable public company. Microsoft was valued at $2.89trn as markets closed on Friday; Apple at $2.87trn. The iPhone-maker has been the most valuable listed firm for most of the past 12 years. But Microsoft’s investments in artificial intelligence have helped it catch up. The software giant last closed above Apple in 2021. — The Economist (Jan 13, 2024)

Thu, 14 Dec 2023 23:22:54 -0500

It was fun to be able to speak at the Mntech Tech Outlook 2024 event today. Dan Abdul, Sarah Engstrom, Eric Hanson, and I introduced our companies and covered a lot of ground with Jeff Tollefson moderating. Topics ranged from GenAI, talent development, technology strategy, and more.

OpenAI Rewind: New Board, Altman Returns ⏪

Wed, 22 Nov 2023 01:46:27 -0500

Wow. First OpenAI…

We have reached an agreement in principle for Sam Altman to return to OpenAI as CEO with a new initial board of Bret Taylor (Chair), Larry Summers, and Adam D’Angelo. — OpenAI

Then Sam Altman…

i love openai, and everything i’ve done over the past few days has been in service of keeping this team and its mission together. when i decided to join msft on sun evening, it was clear that was the best path for me and the team. with the new board and w satya’s support, i’m looking forward to returning to openai, and building on our strong partnership with msft. — Sam Altman

Then Satya Nadella…

We are encouraged by the changes to the OpenAI board. We believe this is a first essential step on a path to more stable, well-informed, and effective governance. Sam, Greg, and I have talked and agreed they have a key role to play along with the OAI leadership team in ensuring OAI continues to thrive and build on its mission. We look forward to building on our strong partnership and delivering the value of this next generation of AI to our customers and partners. — Satya Nadella

All messages posted in a 6 minute period. Is Humpty Dumpty together again?

DALL-E rendition of Satya Nadella and Sam Altman celebrating after successfully putting Humpty Dumpty back together.

If you want to read more than you really ever could justify reading about this.

Prediction: Microsoft will acquire OpenAI

Tue, 21 Nov 2023 08:08:48 -0500

I’ve been following the self-created chaos at OpenAI since Friday’s shocking announcement that the OpenAI board fired Sam Altman. I thought it was a made up headline at first, probably AI generated. My son asked me what I thought would happen and there were three paths I felt were likely.

Every venture firm in Silicon Valley would be offering blank checks to Altman to create a new AI company not limited by the non-profit aspects of OpenAI.
Satya Nadella at Microsoft would offer Altman to join Microsoft and lead all things AI with an unlimited investment budget.
OpenAI would hire Altman back because they can’t move forward without him.

On Sunday night Nadella announced Altman was joining Microsoft, along with others. Option 2 seems to be the choice.

However, now we know that Altman hasn’t signed any agreement with Microsoft. We also know that 700 of 770 employees at OpenAI have requested the board to resign and for Altman to return.

The OpenAI board seems to be fairly clueless here. With the firing of Altman, they effectively pushed Humpty Dumpty off the wall and broke OpenAI into pieces. I suspect that right now Nadella and Altman are the ones trying to figure out how to put it back together again. And I suspect there is only one real option:

Prediction: Microsoft will acquire OpenAI.

Altman can’t be two places¹.
Microsoft is a big investor and partner in OpenAI.
There would be collateral damage to letting OpenAI fail.

In the end Nadella wins in every dimension here. He ends up being the adult in the room putting Humpty Dumpty back together. Altman continues as the visionary AI leader he is. OpenAI remains as a strongly recognized brand. The OpenAI board is dismissed and noted for smashing the bottle after they captured lightning in it.²

Yes Jack Dorsey led Square and Twitter and Elon Musk leads multiple companies. I’m doubtful Altman ends with two roles, leading OpenAI and at Microsoft? Seems unlikely but is possible. ↩︎
OpenAI’s rather unique governance structure with a non-profit at the heart of it makes this more complicated, but I don’t think impossible. ↩︎

Thu, 16 Nov 2023 17:56:32 -0500

We had an incredible day at the Microsoft Technology Center learning and exploring the many AI offerings we can use, and much more. Impressive stuff that left my mind spinning.

Thu, 16 Nov 2023 17:48:08 -0500

This AI display at the Microsoft Technology Center is a great way to show real-time AI capabilities. It showed picture description, object identification, speech recognition, and language translation in real time.

Sun, 12 Nov 2023 22:23:34 -0500

I created my very first custom GPT tonight — GTD Coach. This GPT is designed to help practitioners of GTD that use OmniFocus improve their practice and get advice. I started with more robust questions around different levels of planning and gave great suggestions.

Thu, 02 Nov 2023 12:51:26 -0500

Speaking on a panel at Minneapolis CIO Summit this afternoon — Generative AI: Will it Revolutionize Customer and Employee Experiences? Looking forward to sharing some of the wins we’ve seen.

Update: See LinkedIn post with photos of panel.

Fri, 29 Sep 2023 08:57:24 -0500

Starting AI for Business conference at Carlson School of Management, Management Information Systems Research Center (MISRC). Looking forward to a focused day of learning on one of the most critical topics of the moment.

Tue, 26 Sep 2023 21:54:35 -0500

Seems like $4 billion is the minimum cost to Amazon for missing the LLM wave.

Amazon will invest up to $4bn in Anthropic, an American AI startup founded by former employees of OpenAI, who have built a chatbotcalled Claude. Amazon will acquire a minority stake in the company, which was valued at $5bn earlier this year, to better compete in generative AI. Anthropic will work closely with Amazon Web Services, the tech giant’s cloud-computing arm. — Economist Espresso, Sep 25, 2023

Wed, 19 Jul 2023 14:07:40 -0500

I think there is a huge potential for Generative AI in tools, but this is a heavy expense that is going to make it a tough decision.

Microsoft announced that it would charge $30 a month for AI features in its productivity software. That could mean a cost increase of up to 83% for the millions of subscribers to products such as Word and Excel. Shares in the firm jumped to a record high at the prospect. — Economist Espresso, July 19, 2023

🤔

"Bison & Bitcoin"

Thu, 25 May 2023 20:41:00 -0500

I recently threw the term “Bison & Bitcoin” out with some friends. I love bison. 🦬 The name kept bouncing back to me and I thought that would be a fun newsletter. Generative AI is an amazing brainstorming partner, so for kicks I asked GPT-4 to elaborate and DALL-E to make a cover image. 😂

Prompt: Create a description for a newsletter called “Bison & Bitcoin”. The newsletter has primary themes of innovation, sustainability, and self-reliance, and highlights the shared properties of resilience and strength that Bison and Bitcoin share.

“Bison & Bitcoin” is a remarkable newsletter that transcends traditional boundaries, blending the worlds of ecology and finance through the lens of innovation, sustainability, and self-reliance. It’s an exploration of the innate resilience and strength shared by the American Bison, an enduring symbol of the wilderness, and Bitcoin, the pioneering force in cryptocurrency.

This newsletter is published bi-weekly, each edition shedding light on the enduring strength and adaptability of these two entities. For the Bison, readers get to delve into the animal’s natural history, their journey from near-extinction to resurgence, and the significance of their role in promoting sustainable ecosystems. This deep dive into the life of Bison underscores themes of resilience, offering inspiring insights into nature’s ability to recover and thrive.

Simultaneously, the Bitcoin section dissects the resilience of this digital currency amidst the volatile economic landscape. It explores the trajectory of Bitcoin’s development, its role in driving financial innovation, and how it embodies self-reliance in the increasingly digital economy. It takes readers through market trends, tech advancements, regulatory changes, and how Bitcoin continues to thrive despite these challenges.

By drawing intriguing parallels between Bison and Bitcoin, the newsletter encourages readers to perceive resilience and strength in a new light, and apply these principles in their own lives for a sustainable future. Ultimately, “Bison & Bitcoin” is not just a newsletter; it’s a bi-weekly exploration of resilience in the natural and financial worlds, providing compelling content that provokes thought and inspires action.

Tue, 23 May 2023 21:42:00 -0500

Another example of generated content fooling people. We are going to see a massive acceleration of this. We need to start signing content to have any hope of proving it is real.

A fake image showing an explosion near the Pentagon went viral on Twitter, triggering a brief dip in stockmarkets. The picture, which appears to have been generated by AI, was shared by several verified accounts on the social-media platform including one that falsely claimed to be associated with Bloomberg News. The image was probably tweeted first by an account promoting the QAnon conspiracy theory. — The Economist Espresso on May 23, 2023

Mon, 01 May 2023 21:29:00 -0500

A thesis I’ve pondered for years and won’t go away…

The real “threat” to capitalism isn’t socialism or communism. The thing that may break capitalism is technology.

Give me a place to stand, and a lever long enough, and I will move the world. — Archimedes

Technology is leverage. Technology allows a small number of people to create immense output. The potential present in AI will allow new kinds of output that were never before possible, in some cases with no people at all. Technology has brought efficiency improvements for as long as it has existed, and we should assume that will continue.

With so much leverage the labor market fails. There is no market to exchange your abilities for. There may be no buyer for human capacity. It isn’t a question of retraining if there isn’t anything to train to.

Thu, 27 Apr 2023 09:01:38 -0500

Conjecture: 🤔

As AI makes the Web more synthetic, Crypto will make it more human.

Sun, 16 Apr 2023 06:58:59 -0500

For a technologist, I can think of no better way to spend $20 than a month of experimentation with ChatGPT Plus.

For another $20 the second best is to get a little Ethereum and DAI, create a wallet, and learn how to buy a $1 NFT on NiftyInk.

MnTech Leadership Forum: Generative AI

Thu, 23 Feb 2023 19:28:00 -0500

Today I hosted the MnTech Q1 2023 Tech Leadership Forum. The organization is changing the format of this event to have technology leaders run these quarterly forums. I thought it would be fun to focus on the topic that is making so much press lately — Generative AI: Harnessing Disruption into Innovation.

I was lucky enough to get three others to join in the dialog:

Todd Van Nurden, Principal Developer Cloud Architect and Mad Scientist, Microsoft Technology Center’s
Sam Pierson, Chief Technology Officer, Talend
Tammylynne Jonas, Global CIOGlobal CIO, Donaldson

Some of the topics discussed:

Why now? AI has been in the works for a long time. Has there been a technological breakthrough recently?
How does Generative AI affect how we build software.
How does Generative AI benefit from these early releases.
Opportunities technologists see to use this capability in their organizations.
What are some of the risks to keep an eye on?
AI as a coach or mentor?

Since it was an invite only event for members there isn’t a video to share. 😕

Of course there is a POAP for the event! And even more exciting, the POAP is also held by POAP.MnTech.eth which is the first POAP that MnTech is officially producing.

Sat, 14 Jan 2023 09:06:51 -0500

Neeva recently introduced AI search results. It works well for something so new. This search is a good example I was doing for Tyler where the AI was great. The references to sources are key.

AWS re:Invent 2017 Keynote Notes

Fri, 01 Dec 2017 19:50:00 -0500

My quick reaction to the notable announcements at AWS re:Invent 2017.

Compute

Amazon Elastic Container Service for Kubernetes - Happy to see AWS supporting Kubernetes natively! This is a first step that should result in a better solution for customers that are trying to navigate through platform decisions.
AWS Fargate: Run Containers without Managing Infrastructure - New offering from AWS that promises to deal with the backend of container orchestration.

Databases

Amazon Aurora Serverless - What is a Serverless database? I’m not sure I know, but the idea of a database that has no cost other than storage when it’s idle is pretty amazing for certain workloads. Sadly this is only available for MySQL at the moment, but they promise PostgreSQL coming soon.
S3 Select and Glacier Select: Retrieving Subsets of Objects - Wow, this is pretty cool. Take a bunch of data and dump it into an S3 bucket and then execute SQL-like select commands against that data. This enables some very interesting capabilities.
Amazon Neptune: A Fully Managed Graph Database Service - Graph databases are a bear to pick, use and run. AWS offering a cloud hosted graph database will hopefully address all of these issues. There are a large number of use cases where graph databases make a lot of sense but so many people have been burned by them that they tend to solve a graph problem using a document or relational database. It’s cool to see that this supports RDF and SPARQL, core concepts of the Semantic Web. It also supports TinkerPop3 (really? can we get better names?).

Machine Learning

Amazon SageMaker: Accelerating Machine Learning - Picking models for machine learning training is an often overlooked complexity with doing this right. This bundles that into one tool and should make it easier. We’ll see.
Amazon Comprehend: Continuously Trained Natural Language Processing - I’m sure this has some great use cases around sentiment analysis but the surveillance capabilities creep me out.
Amazon Transcribe: Accurate Speech To Text At Scale - It just got significantly easier to add speech-to-text capabilities to various applications.

Other

Announcing Alexa for Business: Using Amazon Alexa’s Voice Enabled Devices for Workplaces - I actually bought an Echo for us to try and use in our operations area. Idea was to make some read-only data available via voice. It didn’t go anywhere in large part because the security side was too complicated. Theoretically this would solve that, but I’m pretty bearish that this offering is something that many folks would care about.

Sat, 12 Mar 2016 06:43:47 -0500

Alpha Go — “Machines have conquered the last games. Now comes the real world.”

After third straight win, Google's AI claims victory in historic match with Go grandmaster Lee Sedol: https://t.co/aCfdQ8f3uH
— Cade Metz (@CadeMetz) March 12, 2016

Thu, 10 Mar 2016 08:01:13 -0500

Half of me is excited about AlphaGo playing so well. The other half of me wishes it wasn’t.

Wed, 09 Mar 2016 22:40:41 -0500

Good luck to Lee Sedol in match 2 v Alpha Go!

I wish I could say I was confident about winning match 2. One can but try.
— Lee Sedol (@lee_sedol) March 10, 2016

Slapped with a Large Trout

Mon, 04 Nov 2013 01:00:00 -0500

A funny thing happened on Twitter yesterday. My wiki project that keeps a database of nutrition facts Tweets every 6 hours with a random food from its 28,000+ entries. Yesterday, it highlighted trout:

Nutrition facts for Fish, Trout, Mixed Species, Raw http://t.co/8XmR9NlCrl #food #random
— Wikinosh (@Wikinosh) November 4, 2013

Which promptly received an IRC-style trout slap!

Tweet from The IRC Trout has been deleted (theirctrout / 397243070736904192).

So great. Two bots being surprisingly funny. Trout slapping is a meme that started the 1990’s in IRC. It warms my heart to see it alive and thriving on the modern web. 😀

AI on Jamie Thingelstad

The nnxyz rooms

WT353 — Release Notes for Every Day

Weekly Thing Archive Improved

mb 2.0 with MCP support

Weekly Thing Podcast Improvements

Building and Puzzling about Software

So what about programming?

Where does this leave us

Elixir's Agentic Product Team

Thingy Evolved

The Weekly Thing Team

Massive Weekly Thing Website Build

Your Version Number: The Website

Your Version Number

Your Version Number: Work Edition

Hypergrowth

Software Is Liquid

Where we are

Getting philosophical: software is becoming liquid

Every paradigm built on “software is expensive” needs re-examination

Managers belong in the code

Rethinking velocity

The change curve, and the rate of change

Automatic Programming Speed

Making Elixir Smarter

MnTech Outlook 2026

Tron: Ares

Use Apple Intelligence Models in Shortcuts

SPS TechJam 2025

Blog Categories via LLM

Category Introductions via LLM

Writing is Thinking

Emerged from water

Vibe Coding POAP2RSS

Vibe Prototyping

Expand Features

Create Service

Get Professional

Build the website!

Good Vibes!

Are We Merely Machines?

How I Use AI in the Weekly Thing

Subject

Fortune

Byline

Supporting Members

Overall Editing

Weekly Thing GPT

A computer can never be held accountable

MISRC Digital Leadership Conference

Two AI Bots Discussing the Weekly Thing

292 / Signal, Checkboxes, Hope

287 / Plinky, Piccolo, Privacy

276 / Contextual, Copilot, Collections

Multiple Issues

MnTech Tech Connect 2024

Minnebar 18

OpenAI Rewind: New Board, Altman Returns ⏪

Related Links

Prediction: Microsoft will acquire OpenAI

"Bison & Bitcoin"

MnTech Leadership Forum: Generative AI

AWS re:Invent 2017 Keynote Notes

Compute

Databases

Machine Learning

Other

Slapped with a Large Trout