Journal 2025-06-06

Weightless Thought

Ok, Vibe Coding proper, that's mostly hands free on purpose. But recently I've done a fair bit where I've had a fairly solid plan (even if it isn't written down), just let the LLMs have at selected parts, in some vague approximation of systematic order. Not a dissimilar situation to working with other people. I've only very occasionally had a management role IRL, but have collaborated on loads of projects of different sizes and shapes over the years. To move forward you have to have trust in your colleagues. Ok, you occasionally encounter arseholes with some kind of destructive contrariness, or folks with more pressing issues on their mind. Through no fault of their own, they aren't fitting the role you expect. Fair enough, you get used to those kinds of unreliability, work around as necessary. You make allowances, and from a personal standpoint do your best to mitigate disaster, avoid winding up the proverbial río de mierda, senza paddle.

Then there's AI. I think it would be fair to say that although it lacks a sense of responsibility, accountability in the usual sense, it can evaluate and judge and make decisions. It's entirely on us how we project those things into practice. We are the cybernauts, we have our hand on the tiller.

I'm rambling right now because on the other screen I have Windsurf AI (their own custom VS Code build, plus a pretty good model that's free right now), ploughing away writing tests. I need to keep checking, easiest being Source Control tab, that it's not straying where I don't want it. I'm typing during a thoughtless wait. Not to be confused with thoughtless weight, that was the boss I had in my first proper job (long) after leaving school. He was a self-important, what passes for fat cat in Buxton, arsehole.

Aren't they good at making things up!!?

There's gaslighting and AI gaslighting. The other day I had a big push forward with #:semem, rapid progress. Too rapid. I was well aware that most of the material that Claude or whoever was producing would mostly only serve as placeholders, I'd have to go back over everything systematically. One chunk did get under my radar. A whole bunch of material I thought was getting persisted through SPARQL was in fact being looked after by the graphology JS lib. I had to look it up. But I've been careful to keep most things loosely coupled, so rather than revert-backtracking I can refactor. But still.

What prompted me to mention all this is I just realised that a feeling I get from working from these agents is much like that I've had with some of the things I've experienced in the past. A concrete scenario would be visiting a country with a culture that's far removed from what I'm used to (which, for someone that grew up in rural Derbyshire, is most). But a closer parallel would be some of the things I've tried studying. The things that take a suspension of disbelief. Ok, this test-building is taking too long, I'll have to intervene. So I'll dead-end the train of thought here by mentioning one of the more notable things I've tried studying (several times) : Category Theory. Yeah, that's what I mean by Weightless Thought. When things are seriously untethered. Mind-altering substances when you are not used to them; Category Theory; AI-assisted dev.