Our workflow always looks like this:
1. Split our tasks into small sub tasks 2. Write a prompt for each sub task 3. Wait for AI to complete 4. Verify 5. Re-prompt if the verification is not meeting our expectations, if it meets our expectations, give the next task
Most of the time is spent waiting and checking progress manually. If I step away from the computer, I lose visibility.
So I am exploring to create, an AI developer assistant that lives in Slack, Discord, and mobile. Developers can assign these small tasks with expected result to the assistant and while they sleep, the assistant picks them up, runs them through tools like Claude Code and creates pull requests. When the team wakes up, they can review PRs or ask for rework even from mobile.
Questions - 1. Would this be useful to your team? 2. What concerns or blockers would you have before trusting an AI assistant like this?
Any feedback on this would help a lot.
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.