I save every prompt and associated ai-generated diff in a markdown file for a steganography package I'm working on.
Check out this document: https://github.com/sutt/innocuous/blob/master/docs/dev-summa...
In particular, under v0.1.0 see `decode-branch.md` prompt and it's associated generated diff which implements memoization for backtracking while performing decoding.
It's a tight PR that fits the existing codebase and works well, you just need a motivating example you can reproduce which can help me you quickly determine if the proposed solution is working. I usually generate 2-3 solutions initially and then filter them quickly based on a test case. And as you can see from the prompt, it's far from well formatted or comprehensive, just "slap dash" listing of potentially relevant information similar to what would be discussed at an informal whiteboard session.
I do not care if engineers on my team are using AI. In fact, I think they should be. This is the new world and we need to get used to it.
But it’s still your work, your responsibility. You still have to own it. No cop outs.
Yeah "consider firing" is a bit of an extreme kneejerk reaction, but I just feel like we HAVE to create consequences for this nonsense.
Junior dev vibe codes some slop to solve a problem that was barely a problem. The "solution" to that not-a-problem was 2500 lines of slop. Unused variables, spaghetti logic, unit tests that were clearly write-once, read-never.
The slop at that point is the meta-problem, the real problem becomes me trying to convince (through multiple rounds of code review) that this was not maintainable code, and not a tenable development model.
All those rounds of review take TIME and mental effort. At some point, when the code review takes more effort than the code owner contributed, the value proposition is shot.
Generate the code however you want (llm, bit twiddling, whatever), but the effort and care must be there. If you can't use an llm skillfully and vouch for the output, it's gonna be a no from me dawg
I call that a good day. I've seen people push 2000 line PRs. The worst was 5000 lines. FML.
I haven't been able to get any of your recently published projects (Scribe, volknut, a few others) running successfully on linux, whether following your documentation, using the tagged releases, or reverse engineering the true functionality and CLI arguments working from your provided source code I found to have wasted my time.
It's difficult to believe you when your own documentation is entirely wrong or 404'd.
I'd genuinely love to see your projects in action, since the concepts sound promising, but you need to back it up with shipping properly.
Understand that I'm one man working on 20 projects simultaneously, with 5+ under active development at any one moment, so release stabilization and cadence will take a little bit to lock in.
This is why AI code review continues to be mostly useless in my experience. If you instruct an LLM to make PR comments it will come up with some, every time, even when none are warranted.