Cognitive debt and AI: from Ruby on Rails to no longer following the code

3 min read #Youtube #Ai #Agenti #Philosophy

Yesterday I went to a small conference in Florence about cognitive debt: the knowledge of what the code does that a person or a team accumulates, or rather loses along the way.

Every step of abstraction is cognitive debt

I take the concept further. There is cognitive debt every time you climb a step of abstraction.

The mechanic of the old days machined the parts on a lathe. Today’s mechanic removes and refits the manufacturer’s parts. The electrician doesn’t repair the boiler’s board: they pull it out and put in a new one. Have they lost knowledge? Yes. Do they work worse? No, they work at a different level.

In programming: how deep do you go?

Programming is the same. There’s the language, below it the design patterns, below that how the code is compiled, and at the very bottom how it ends up in the processor’s registers. You can go all the way down, it’s all there.

But do you really need to know all of it? It depends on the context. Someone writing a driver has to go down to the registers, someone building a management app doesn’t.

Ruby on Rails and pragmatism

Until six months ago I was writing Ruby on Rails by hand. Rails is the framework of total pragmatism: development hours were the most expensive item in a project and Rails cut them. Convention over configuration, less code, straight to the point.

AI cuts the cost of code by 10 to 100 times

Now AI arrives and that cost drops by 10 to 100 times. It’s not an incremental improvement, it’s a change of step.

And AI goes so fast that you can no longer keep up with the code. In my opinion you no longer have to. Frontier models, American, Chinese, even the heavier open weights ones, write better than the vast majority of developers. Period.

The developer becomes product owner and architect

Our job shifts. Today the developer is a product owner and an architect: they know what’s going on underneath in terms of architecture, performance and scalability, delegate the code to agents and check the critical parts.

And the most critical part of all is the data. Code can be rewritten, lost or corrupted data can’t. That’s where attention has to stay very high: schema, migrations, tests.

I also talked about this while programming with agents during a walk.

Those who don’t convert

Those who want to keep writing code by hand are, in my opinion, living on borrowed time. It’s a harsh judgement, I know, but it’s the same one you could have made about those who wanted to keep machining parts by hand.

That’s what I think. How deep do you go? Tell me in the comments on the video (it’s in Italian).