Ideas, opinions and experiments
Best story wins
We train AI agents with drive and no way to give up. What motivation research says about where that leads, why a half-wired prototype is the small version of the same thing, and what we lose when machines start telling our stories.
Read more
Rules, judgement and the noise in between
Everyone talks about AI bias, far fewer about AI noise. What Kahneman's Noise teaches about rules vs judgement, where new decision models like Jev and Laya fit between code and chat, …
Read more
Onboarding onto a codebase nobody wrote
Developers already spend around 58% of their working time reading code and working out what it does. Inheriting a vibe-coded prototype means doing that with nobody to ask, so we …
Read more
Why we write our own skills
A skill is packaged instructions your agent follows on every task it touches, so installing one you haven't read means running someone else's process on your codebase. Why we write our …
Read more
AI writes the code. Experience writes the ticket.
Give an agent a vague ticket and you get plausible code that closes it. Give the same model a real specification and it produces code that ships. A real opening ticket from a live …
Read more
We accidentally built our own version of Jira’s AI …
Atlassian rebuilt Jira's developer pitch around AI agents, and we read the page with a strange feeling of recognition. We've been running that workflow on a production build since …
Read more
We gave an AI agent its own account in our project tracker
An AI coding agent sits in our assignee dropdown between two human engineers. Most of what we learned came from the preparation: the tag that gates the work, the review column it has …
Read more
The demo gap
Somebody on your team built a working version over a weekend, and the board has already seen it. What stands between that demo and a system is most of the work, and it happens to be …
Read more
The convergence is in the reasoning, not in the learning
Two things are actually broken in this generation of AI, and reasoning isn't one of them. A single training run costs the lifetime energy of roughly 3,500 brains, and a five-year-old …
Read more
Why I was quiet on the AI topic, and what I got wrong
I posted about AI and machine learning for over ten years and got about three likes for the trouble. Then it became the biggest topic there is and I went quiet. A recommender system in …
Read more