4 ms·
In my team, it's making one dev more prolific, and everybody else work harder. The most junior dev on my team was tasked with setting up a repo for a new servi
by firefoxd 1y ago
In my team, it's making one dev more prolific, and everybody else work harder.
The most junior dev on my team was tasked with setting up a repo for a new service. The service is not due for many many months so this was an opportunity to learn. What we got was a giant PR with hundreds of new configurations no one has heard of. It's not bad, it's just that we don't know what each conf does. Naturally we asked him to explain or give an overview, he couldn't. Well because he fed the whole thing to an LLM and it spat out the repo. He even had fixes for bugs we didn't know we had in other repos. He didn't know either. But it took the rest of the team digging in to figure out what's going on.
I'm not against using LLM, but now I've added a new step in the process. If anyone makes giant PRs, they'll also have to make a presentation to give everyone an overview. With that in mind, it forces devs to actually read through the code they generate and understand it.
- vkou 1y agoI think the simpler expectation here is the same one you should have for non-AI code. Don't allow giant PRs without a damn good reason for them. Incremental steps, with testing (automated or human) to verify their correctness, that take you from a known-good-to-known-good state.