ionetan

With agents code has gotten really easy to shape, like fabric. If want to build something rigid, you'd want to introduce sturdy architectural structures to stretch out the fabric of agentic code.

Data and its shape is one such rigid structure. It will stay with you long after the code around it has been replaced. You do want to be intentional about it and build intuition for it.

Once you have that in place, you can stretch out a web of agentic fabric across it, and pin it down with tests.

Code is soft nowadays, but data isn't.

Good point. I also think that the training data is different in this regard. It has trained on (mostly) verified, correct code, but when in comes to free text it has trained on a lot of unfounded opinions.

I think it is more about comprehension. It reveals that AI does not actually understand what it does. It gets code right mostly because it is probabilistic and has seen enough sound precedents to land in something that works, but it does not come from an internal comprehension.

AI has a way of using abstract language, sounding impressive and eloquent, but in reality it is mostly incoherent fluff. You can see this in humans too. Some politicians and academics are good at expressing themselves in clever ways but when you actually try to parse what they say it makes no sense what-so-ever.

I think there’s something to be said for distributed systems that work equally well online and offline. I’ve built an issue tracker that uses Git as its transport, with no central broker or service and eventual consistency.

That means agents can collaborate on issues locally without needing network access to an external tracker, which could help with the sandboxing problem. You can then sync when it is deemed safe.

It’s called Epiq on Github.

I read DHH, and started wondering if he is ok :D

Yes. Started off cleaning school toilets. Went to a bootcamp 10 years ago, taught myself the rest, got to sit in fancy offices and drink coffee, but always felt outta place. "You can take the kid out of the ..." and so on. Guess I gotta pick up the broom again.

Yeah, I guess it is quite hard to tell where the line should be drawn, and we probably need to rely on human intuition and experience. One thing that the article left out that I have thought more of afterwards is that the classification would need to be relative to the enterprise as a whole. So if you are building a gimmick app for a birthday party you might just vibe code the whole thing, including its "core", but if you are a building business-critical sales system for a major company, the impact of something going wrong would rate the core at the upper end of the scale.