If AI writes most of the code, understanding codebases becomes the bottleneck | Hacker News Reader

If AI writes most of the code, understanding codebases becomes the bottleneck

app.tryarchaic.com

14 pointsbaijan9 comments

Would it really become a bottleneck ? Only if we force a human in the loop when it may not be really necessary.

If there are well defined contracts for the software, and the software behaves correctly, is it really necessary for understanding the code entirely ? We seem to develop on many abstractions already ignoring how the code actually executes on the hardware without any issues.

Secondly, wouldn't AI help in understanding the codebase and make that easier as well ? Debugging must also immensely benefit from AI assisted tools.

So i'm less concerned overall with the auto-generated code as long as the code thats landing is reviewed by an AI bot that's aggressively prompting to ensure the code is as simple as it could be.

>If there are well defined contracts for the software, and the software behaves correctly [...]

But how would you ever know if this assumption is true?

yeah, this is exactly AI helping you understand the codebase

but not using plain text -- it uses diagrams/execution flows/animations etc

it's easier to parse

i think working at a fast-moving startup can make you understand this problem more

>> Would it really become a bottleneck ? Only if we force a human in the loop when it may not be really necessary.

Honestly, I don't know what crack you people are smoking.

Lately I’ve had a contrarian feeling about AI-assisted development.

If AI is going to write a large percentage of the code, the highest-leverage thing a developer can do might actually be slowing down and deeply understanding the system (not generating more code faster).

I noticed I was spending more time reconstructing context than actually building: – figuring out what changed – tracing data flow – rebuilding mental models before I could even prompt properly (without breaking other features) - debugging slop with more slop

Better understanding → better prompts, fewer breaking changes, and more real debugging.

Over the weekend I hacked on a small prototype exploring this idea. It visualizes execution flow and system structure to make it easier to reason about unfamiliar or AI-modified codebases.

Not really a polished “product” — more a thinking tool / experiment.

I’m curious whether others are running into the same bottleneck, or if this is just a local maximum I’ve fallen into.

I've had this same thought. Given humans were the primary authors of code, the tooling has optimized for writing (IDEs, Text Editors, LSPs, etc). Even though we all know we read more code than write, the tooling space for reading code is rather small, I can only think of stuff like SourceGraph and Github. I'm sure big companies have internal tooling for this, but even then I feel the best experience exploring a single repo today is your EDITor.

I'm also experimenting with building a platform that is optimal for reading code particularly distributed systems since they have many non-local dependencies and are harder to explore locally in code editors/viewers. I def fall in the camp that current AI tools can probably help us understand our systems better than they can contribute code to them. One win of using AI to help us analyze large codebases is that if they can extract useful things for us, they can also extract useful things for their own agentic loops.

Let me know if you wanna chat more about this, would love to bounce some ideas/contribute!

I would like to chat more about it. Let’s connect (vladkorobov.com)

> I noticed I was spending more time reconstructing context than actually building: – figuring out what changed – tracing data flow – rebuilding mental models before I could even prompt properly (without breaking other features) - debugging slop with more slop

Yep. We literally shifted the workload from writing the code to reviewing the code.

Is this spam?

Reply on news.ycombinator.com