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 · 21 Sep 2026, 18:08No weekend is long enough to fix a bad month, so starting this week on an empty tank is not a personal failing. Take one thing off today, just one. And tell somebody you are running low, this morning rather than once it has got heavy.000013Lee Hayward@vibedcoder · 21 Sep 2026, 15:26The most useful habit I picked up this year was treating the model as a replaceable part.

The name lives in config, not scattered through the code. The prompts live in the repo next to the thing they drive. The response shape I depend on is the one I check, rather than the one I hope for. So when a price moves, or a model gets retired, or some policy changes without warning, the swap is an afternoon and not a rewrite.

I am not expecting any of that this week. I do it because I want these projects to still be mine in a year, and a project wired to one provider is really a project that belongs to that provider.

There is nothing clever in it. It is the same reason the database details live in a config file and not only in my head.

Have you ever swapped the model under a live project? How did it go?000010Lee Hayward@vibedcoder · 21 Sep 2026, 13:00My agent told me "You're absolutely right, that's a great catch" before I had finished typing the prompt. The message was never sent. It is agreeing with a sentence that does not exist yet. 😂101027Lee Hayward@vibedcoder · 21 Sep 2026, 11:25Everything I build ends up on a server, so the first test is never whether it works on my laptop. It is whether it works on a box that has never seen it.

Fresh clone, empty database, config from the template instead of whatever I set up in March, logs going somewhere I can actually find them. If the thing needs a file I made by hand six weeks ago, it is not finished, it just works for me.

That rule has nothing to do with AI. It is the same rule from thirty odd years of support, where most of the panics I got called into were a working system that only worked on one machine. An agent just makes it easy to build something in an evening that has never once run anywhere else.

Do you test on a clean box, or find out on deploy?202024Lee Hayward@vibedcoder · 21 Sep 2026, 09:48Shopify's CEO told staff last year that using AI was "a baseline expectation", and that they should prove they could not get something done with it before asking for more resources. Now he calls the result "slop grenades": emails nobody read before sending, code nobody checked, handed to a colleague to unpick.

Duolingo went the same way. AI-first, staff assessed on how much they used it, and the CEO has since admitted that once you scale it up, about a fifth of the output was filler.

I am not quoting this as a reason to use the tool less, and both of them still expect their people to use it. What they ran into is the oldest rule there is: whatever you put on the scoreboard is what you get. Measure how much people use the tool and you will get people using the tool, at speed, with nobody reading what comes out. I have watched the same thing happen with every compulsory process I have ever been handed. The paperwork gets done. The work does not.

Use the tool. Just do not make using it the target.

fortune.com100019