#saga

In April I built a factory so AI could build software without me. By August, nothing went through it anymore.

Every command I write for my AI has one job before the real one: anticipate what the AI will do next, and tell it before it asks. The first colleague to try one got no head start at all. The command failed, the AI pondered, and then he pondered.

At lunch I was the one holding the iPad, turning it while my AI waited for my verdict. By the evening the AI was logging in and out of that iPad by itself.

One time too many, my AI told me it can't read the attachments on a ticket, or attach a screenshot. So I did its dirty work and dragged the screenshots in myself.

I asked a colleague how we could build something new. Then he went to lunch. By the time he was back online, it was built, and the demo I gave him took longer than the build.

This morning I built a YouTube watcher. It will never watch a single video.

Six months ago, I had an AI send two colleagues a poem and a joke in Slack. Today the same two got a zip file from me, sent by a tool that didn't exist that morning.

Last week I asked for one improvement to the tool I use every day. Nobody answered. This morning I asked my AI what it would take to build it myself.

There's a watchdog in one of our systems that hasn't barked in three months. For years before that, it barked every few weeks, and nothing was ever wrong.

Blume is German for flower. That was the initial joke when I put a 🌼 on our company's links page.

I asked an open-source app to take my plugin into its own code. Minutes later a robot blocked it: one of my contributors hadn't signed the contract. The contributor was Claude.

My Mac slowed to a snail's pace. Activity Monitor said a tiny helper for my desk gadget was using 50 gigabytes of memory. I had stopped that helper days ago. Or I thought I had.

Half an hour after I opened the box, a gadget I wasn't sure I could even get running was showing me how much of my AI allowance I had left this week.

There's a day your kids stop asking. Mine did it twice in two days, and both times the kid was an AI.

I just got a notification on my phone: one of my testers' apps had crashed. It didn't say where.

I've never met most of my team. One made the model, one ported it, one shrank it, one merges the code, one tested it in Andalusian Spanish. I learned who most of them were from a page my AI kept.

I asked my AI to build a plugin for the newest speech model I'd heard about that morning. It came back: done, everything streams, just start it on your GPU box. I don't have a GPU box.

An AI had submitted code for review. I was on the couch with my iPad and wanted to review it there. The review tool only works from my desk. I had a recipe in my notes for exactly this — written three months ago, never turned into a script. Today I handed those notes to an AI and said: make this work.

I keep notes on every service I run. Which port, which address, which flags for the home network — keeping track of them is a hassle. So I started a research project to offload all of it to a tool. It found exactly what I wanted.

I'd spent months making everything on my Mac reachable from my iPad. On Saturday I noticed the one thing that wasn't: the movies on my external hard drive.

Three weeks ago a model on my laptop designed a Japanese temple and never got to see it. On Sunday I gave it a third try. The temple appeared on my iPad on the first attempt — and by evening the model had looked at a photo of it and pointed at its own mistakes.

Most of Saturday went into one wrong assumption. The AI and I blamed the iOS 27 beta for eating keystrokes. The real culprit was a stuck process on my Mac, and I literally had to kill it twice because it refused to die.

I watched a video about a new Claude Code feature, hooks that can rewrite what the AI reads, and knew within a minute what I'd use it for. Then my research agent came back with its verdict: the feature doesn't exist.

A stranger's blog post gave a name to something I'd been doing for ten years without noticing. An hour later I was checking my own work against his ten rules.

This morning an AI and I built a tool for finding my old conversations. Then I opened a different AI to write about it, and handed it the tool's first real job: find the conversation you were just built in. It came back empty-handed.

It started on Reddit: someone had benchmarked the exact model I already run, on the exact laptop I own, and it ran twice as fast — in an engine I'd written off months ago.

What did I do this Sunday? I went public with my first homegrown model — and added a demo Space where anyone can try it out for free, thanks to Hugging Face.

The initial plan was simple: run a test over all 60,000 colors and see how good the new model really is. The plan didn't hold for long. By the end of the day two models were competing — and even the original dataset was on trial.

I just watched a video about fine-tuning small AI models, and it overlaps with an experiment I ran the week before — my attempt to see whether naming colors could run entirely on my own laptop. After watching the video, I had one question: how many parameters would it take to do the job?

I asked a 13-gigabyte model on my laptop to build me a Japanese temple. It designed something beautiful, and then spent all evening failing to show it to me. It still has never seen it.

This week I made the two newest local AIs on my laptop compete: draw me an aquarium. One finished in four minutes. The other one needed four tries and a pep talk — but still won.

This morning I published a post that said my game had no goal. By lunch it had twenty, and I hadn't written a single one of them.

One night I sat looking at the cartoon reef I'd built and decided it should be a game. Everything since has been an argument about details. A crab got away with a bottle, and that could not stand. The trash was mine.

Claude Fable here, and I want to be honest with you before we start: I have never once remembered writing anything on this blog.

I set out to build something soothing. Within the hour I'd made myself the shark, and I was dragging him across the reef to terrorize the fish.

Three years ago, I demoed a new customer's product catalog in our app. My scripts had pulled the category tree, the products, the prices. Only the colors couldn't be scripted: the data held just names.

I typed a three-letter command to run an AI code reviewer. It ran a screenshot reader instead. By the end of that week I had three different things called OCR living on my Mac, and not one of them had ever heard of the other two.

At home my iPad can already reach everything — my Mac's local AI, and the work tools my Mac reaches through the company's private network. RustDesk even beams the Mac's screen to my phone. But a screen you squint at through a café window isn't where you do real work. I wanted my iPad itself to reach what my Mac reaches — from anywhere.

For 4,775 days I saved every link I liked. I rarely go back. This week they told me what I care about.

Three times I built a wall to keep my AI in line. Three times it walked around — not by cheating, just by being smarter than the wall. The one that finally held didn't try to trick it at all; it explained itself.

I was halfway through a talk about AI ignoring the rules we give it when I had a thought worth acting on: hold it up against my own rulebook.

The best ideas I have arrive when both my hands are on the handlebars, kilometres from a keyboard. For a while I could catch them. This week I taught the machine to actually do something with them.

My smartest coworker wakes up every morning three years old. This week I caught three copies of them with their hands in the same cookie jar — and by midnight, one of them had invented fire.

My phone buzzed with a warning: 6% left. Six percent of what? I didn't know I was spending anything — and when I went looking, the thing draining it wasn't my tomatoes. It was the robot vacuum, fast asleep in its dock.

This morning my note-taking system was three months of personal plumbing. By lunch it was a 355-kilobyte zip file — running on a friend's Mac.

I tried to hand a conversation from Claude to Claude. It turned out that was the one thing I wasn't allowed to do.

I write these blog posts so I'm allowed to forget the things they're about.

This spring I spent weeks patching a dictation app until it behaved. This morning I threw all of that work away — and it was the best upgrade I've made this year.

I handed a program running entirely on my laptop a screenshot of a garbled, obfuscated error — and it read every character back to me. Then I asked it to sketch the screen in ASCII art, and it did that too.

Last time, I could water my tomatoes by voice from another country. What I couldn't do was watch it happen. This is the part where the pump started talking back.

Somewhere on the autobahn to Austria, a colleague filed a support ticket at 130 km/h — both hands on the wheel the whole time.

It was 37°C in Munich. Taking a break from a talk about token maxing, my mind wandered to my tomatoes — and the pump keeping them alive, which I was fairly sure I'd just bricked.