Eyelock
Bio coming soon!
Site Rejig.
Carried out some back end rejigging of the site, turned off the comments as they were going to start showing adverts. Let me know if you spot any problems.
In August 2025 I was shouting at an LLM in all-caps
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
KickoffMANDATORY: You MUST work ONLY in the specified working directory. NO EXCEPTIONS.That is Clause 27 of a legal code with a jurisdiction of one machine and no court of appeal. I wrot
Anatomy of a naive prompt library
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Kickoff My 2025 prompt library had a filing system a Victorian clerk would have admired, in service of a legal code no court would recognize. 001-Central-Context.md. `200-Standards-Context.
My first agent managed my music downloads. It still does.
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Kickoff The first genuinely useful thing my AI setup ever did was tidy my downloaded music. Not a rule. Not a threat. A chore, quietly delegated. August 2025, one day after starting my prompt
My best failure took 36 hours
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Kickoff Seventeen commits. Thirty-six hours. Abandoned. Let me tell you about the best project I ever killed. Late January 2026, I took one more run at an old dream: a proper prompt distribut
Recap: Act I — The Before Times
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Three posts in, and the prompt era is behind us. Before the story speeds up, the recap — and a clean place to join if you're new here. Act I was August to November 2025, and it looked like this:
I started logging everything the LLM did — inside git
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Kickoff Three days after Christmas 2025, I started building a system whose entire job was to write down what the AI actually did. Not what I asked it to do. Anyone can write down a request; m
The most important refactor I did all year fit in one line
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Kickoffrefactor: align cli and mcp more with zod schemas3rd of January 2026 — a commit message so boring you'd scroll straight past it. It turned out to matter more than anything else
December 2025: the month the bottleneck moved
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Kickoff The new models didn't make my prompts better. They made my prompts irrelevant. When Anthropic's new Sonnet and Opus models landed in December 2025, something qualitative shifted. For
Recap: Act II — The Building Sprint
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Act II was three weeks of December and January in which I stopped writing prompts and started building structure. The recap, before the story gets a name.**[The bottlenec
The weekend I rebuilt what I already knew
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Kickoff The git history says I built it in one weekend: fifty-eight files, ten thousand lines, landed in a single commit on the 2nd of February 2026. The git history is lying, and the lie is t
I built an IDE over dinner
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Kickoff By early January 2026 my agentic setup had given me a brand-new problem: I was too productive to know what I was doing. Ten-plus terminals open at once, most of them running agents on
The day my project renamed itself
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Kickoffrefactor: rename persona → harness across entire codebase**There's a commit in my history dated the 6th of April 2026 that I'd frame and hang on the wall, if you could do that with
Package the environment, not the prompt
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Kickoff Eleven days. That's how long it took after April's persona→harness rename for the idea to jump products. April 6th: ynh renames its whole world around harnesses. April 17th: my IDE gr
Recap: Act III — Naming the Thing
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Act III is where the series earns its title. February to May 2026, recapped — before the story finally walks into the office.**[The weekend rebuild](/off-the-model/08-the
AI Week, four times: nearly forty people, one room, one library
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Kickoff For most of this series I've been alone in a room with git repos, answerable to nobody but the commit history. This post is about the opposite: nearly forty colleagues, four consecutiv
I had to walk backwards into work
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Kickoff At home I had versioned, installable agent harnesses with drift detection and a marketplace. At work, I didn't bring any of it. Instead, I forked a workflow baseline the company alrea
Recap: Act IV — Into the Office
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Act IV was the shortest act and the hardest engineering: people. The recap, before the final stretch.AI Week, four times — four back-to-b
An AI agent filed a bug report on the harness it was running in
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
KickoffAuthor: Claude Code agent (Opus 4.7), driving the Collective harnessOn the 23rd of May 2026, a detailed bug report landed in one of my repos, and that's its author line. Three
The map I wish I'd had
- None ,
- Eyelock
- Technology , Off the model
- July 20, 2026
Kickoff Eleven months. Seven repositories. Four abandoned projects. One rename that mattered. This is the last post of the series, so here's the whole journey on one page — and the map I wish
The harness guides
- None ,
- Eyelock
- Technology , Loosen the leash
- October 8, 2026
Here is the thing we want. You hand a coding agent a task, and before it starts you already know three things: what "done" means, how much it is allowed to spend getting there, and what it cannot do n
The factory decides
- None ,
- Eyelock
- Technology , Loosen the leash
- October 8, 2026
Here is the thing we want. Someone labels a ticket, and a little later there is a draft pull request for it. Nobody typed a command. Every decision on the way is written down where you can read it bac
The factory remembers, and memory never decides
- None ,
- Eyelock
- Technology , Loosen the leash
- October 8, 2026
Here is the thing we want. A factory works through a pile of tickets overnight and some of them fail. In the morning you do not want a count. You want an answer: what failed, with which exit code, wha
The memory remembers
- None ,
- Eyelock
- Technology , Loosen the leash
- October 8, 2026
Here is the thing we want. An agent finishes a task, and what it learned does not vanish with the session. The next session, on my machine or a colleague's, starts knowing that the tests need a clock
Three layers, three refusals
- None ,
- Eyelock
- Technology , Loosen the leash
- October 8, 2026
I asked a coding agent to remember my name. It wrote the fact into its own memory directory. Not the memory I had just installed for it. Its own.That one line of behaviour cost me an archit
Together, then taken apart
- None ,
- Eyelock
- Technology , Loosen the leash
- October 8, 2026
Here is the thing we want. A ticket gets a label and, some minutes later, a draft pull request is waiting for you. Three separate pieces made it: one decided what to run and when, one set the rules an
The factory drives the harness
- None ,
- Eyelock
- Technology , Loosen the leash
- October 8, 2026
The question I used to ask was whether the agent could be talked into something. It can. Anything that reads text can be talked into something, and a ticket is text written by whoever can open an issu