#agent

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 was in bed, on my phone, watching my AI drive a fake phone, to fix the app I was talking to it through.

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.

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.

The last thing the AI told me before my allowance ran out was that it was pushing my page to Confluence. It wasn't. When the allowance came back, I typed \"Continue\", and it did the exact same thing again.

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 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.

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.

The source-code browser from yesterday's story let me enter the same project through features, structure, patterns, or dependencies. The next morning, I pointed the idea at my Obsidian vault. It came back with folders, broken links, and a list of notes that linked to each other. I had turned a way to explore into a plumbing report.

My weekly AI budget hit 99% with one part of the saga still unwritten. The replacement cost nothing, refused to say who built it — and by evening it had its own fan page.

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.

The best idea in my database tool is stolen. I found it in a tool that edits Word documents.

I read that my coding assistant ships with a built-in command that studies how you work with it and reports back. The first section, 'At a Glance', looked fine. The details read like it took five of my three thousand sessions and built the whole thing out of them.

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.

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 watching Matt Pocock's 'AI Coding is exhausting,' and thirty seconds in he hit a nerve I'd been ignoring for weeks.

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.

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.