Everyone seems to be building AI agents right now, but I'm curious how engineering teams are deciding where to use them.
If you inherited a large codebase (say 500k+ lines of code), what would you look for before deciding that a workflow should become an AI agent instead of a traditional service or automation?
Some things I've been thinking about:
Repetitive workflows that engineers perform every day.
Processes involving multiple systems (GitHub, Jira, Slack, databases, etc.).
Well-defined business rules with clear outcomes.
Existing APIs that an agent could interact with.
On the other hand, I feel AI agents are probably a poor fit when:
The process changes every week.
Most knowledge lives in people's heads.
Success criteria aren't clearly defined.
I'm currently exploring this problem and trying to build a framework for identifying "AI-agent opportunities" in existing codebases, but I'd love to hear how others think about it.
How would you evaluate whether a workflow is worth turning into an AI agent?
Top comments (0)