PaulHoule
11 hours ago
All I can think of is
https://video.disney.com/watch/sorcerer-s-apprentice-fantasi...
The core thing about AI coding is “sometimes it works and sometimes it doesn’t”. There are times when I ask an agent to do something and it does a great job, other times when it takes three attempts. There was that time I was struggling with a TypeScript code base where the errors didn’t make sense to either me or the AI but in three days of hacking on it with AI help I whipped the code into shape, understood what was going on, and the AI became able to do application-level tasks (“add a new field to this form”)
I just can’t see an AI coder going 8 hours and making anything that doesn’t take 16 hours to whip into shape.
narayanahari
10 hours ago
Great feedback. Thanks for sharing. Totally Agree. AI coding is inconsistent today. We are seeing the same thing using Claude and Cursor daily. Some tasks work beautifully but others spiral.
Thats the reason i am thinking this tool is to offload small and well-scoped, low risk tasks that developers usually queue anyway. Things like making tiny UI changes, update some validations, write docs, adding more test cases.
I am thinking of this more of a "overnight helper" than a full dev replacement. If this tool can handle 3-4 tasks while we sleep and open clean PRs thats already 20-30% productivity bump for small teams. The goal is just get the continuous motion to increase productivity.
Some usecases that i am thinking 1. Minor or Major support tickets (update error codes, fix messaging, update validations) 2. Tiny UI changes (Change the text color of dark mode primary buttons to black, add more countries to the list) 3. Fix issues reported by static code analyzers 4. Increase test coverage by adding more UTs on these specific areas.
More of an entry level intern who can help in getting grunt work completed while the devs focus on the important ones.