Comment by i_love_retros
7 hours ago
Who's reviewing all the code you are churning out with ai? If everyone is used to maintaining not developing software it doesn't sound like they'd be best suited to have to review lots of complex pull requests.
It sounds like you are moving very fast and probably have people just clicking "approve".
Good luck for the future to who ever owns your company!
When I setup systems, I thoroughly document them, test them, develop them according to architectural best practices. My AI assisted code generation is lightyears ahead of what I see in companies I have worked for. The best they —the companies—-do is hire expensive consultants. Who sell them preconfigured system. And when you look into those systems you won’t believe your eyes either. Because you instantly realise that those devs do not know much about architectural patterns, aout systems design, about software development best practices. Yet they sell their systems as well, because they offer a niche product where they have only a handful competitors.
In that setting someone with solid software engineering background using AI to solve problems is like a wizard from the team‘s perspective.
When I worked for startups I was constantly panicking to miss the latest tech trends, and I feared that I would be not marketable in case I didn’t catch up. But in mature companies things work much slower. They work with decades old technology. In that setting not the latest tech counts but being able to solve problems, with whatever means you can.
did i miss it or did you still not answer
> Who's reviewing all the code you are churning out with ai?
I don’t review every single line that AI generates. I glance over the files to see if they meet my standards, prompt it to rewrite this or that portion, when necessary. Or change it myself.
Writing code is the most tedious part, not reviewing.
2 replies →
He didn’t answer because he didn’t even read your comment. Likely a bot
1 reply →
Gemini of course! /s
Personally I've had mixed experience when I let Sonnet 3.7 document my (and its) code and write commit messages, for some wip stuff it's alright but it soon gets out of hand and because it doesn't really have a direct view in my mind it ends up documenting what it sees instead of the intention behind it, which is totally fair but eh.
So yeah, mileage varies and agentic tools usually spit out more code and redundant comments than I'd like to review. I'm still waiting for a company to develop some sanity check for this somehow but snapshot testing and manual review aren't enough sadly.