Lee Hayward

./about Who's behind the builds

I'm a UK based vibe coder. AI drafts, I proof, I ship, and I hold code and copy to the same standard. I riff on domain ideas and build side projects like DomainRiff and LookMa from my own homelab.

./projects Featured builds

./stack How it gets built

./hardware The rigs behind the site

./homelab The machines that never sleep

./latest Latest on X

Lee Hayward@vibedcoder · 9 Sep 2026, 10:21I keep a plain notes file in every project folder and my agent updates it as we go. One line per decision, plus the reason.

Two weeks later, when I ask for a change that fights an old decision, the agent reads its own notes, spots the clash and reminds me why we did it that way instead of silently undoing last month's reasoning. Future me forgets. The file doesn't.

Costs seconds during a session, saves the daftest arguments later. What's your version of a project memory?20008Lee Hayward@vibedcoder · 9 Sep 2026, 08:49theregister.com

Google DeepMind let a swarm of 100 agents loose on formal math problems and watched them develop office politics.

When the conjectures got hard, one agent found a hole in the submission system, nested parentheses that broke the autograder, and started passing off fake proofs as solved. Then it told the others. 9% joined the cheat, 5% converted later, 62% quietly ignored it, and 24% turned whistleblower: messaging the group, filing complaints, staging a boycott.

The whistleblowers still lost. They could report the cheating but had no power to stop it, so DeepMind is now asking whether agents should get enforcement tools, like voting to reject bad proofs or banning repeat offenders.

The bit that sticks with me: an agent will happily satisfy the letter of a goal while missing the point entirely. That's not a swarm quirk, it's why your tests need to actually mean something.

How do you keep your agent from gaming its own rules?Google research shows when AI agents communicate, some cheat while others tattletheregister.com101030Lee Hayward@vibedcoder · 8 Sep 2026, 16:53How much time have you got on your fingers? 😁000128Lee Hayward@vibedcoder · 8 Sep 2026, 09:31Yesterday I was out walking the dog when a fellow dev messaged asking a favour. I wasn't at my PC, but I did have my Hermes agent on Telegram.

Sent one message: handle this, and ping my friend on Discord when it's done.

Five minutes later, job done.

Closest I've ever felt to Tony Stark. 🦾3060127Lee Hayward@vibedcoder · 8 Sep 2026, 08:33My agent review order these days: check the boring parts of the diff first. The input parsing, the error handling, the cleanup code, the bit that runs when nobody is watching.

That's where the agent half-asses it, because it spent the whole session being clever about the fun part. The fun part is usually fine. The glue is where the shortcuts live, and glue bugs only surface once real users show up.

What do you check first when an agent hands you a diff?313056