DEV Community

Dror Wayne
Dror Wayne

Posted on

๐–๐ก๐š๐ญ ๐ค๐ข๐ง๐ ๐จ๐Ÿ ๐ข๐ฌ๐ฌ๐ฎ๐ž๐ฌ ๐ฌ๐ก๐จ๐ฎ๐ฅ๐ ๐ฒ๐จ๐ฎ ๐๐ž๐ฅ๐ž๐ ๐š๐ญ๐ž ๐ญ๐จ ๐ฒ๐จ๐ฎ๐ซ ๐€๐ˆ ๐œ๐จ๐๐ข๐ง๐  ๐š๐ ๐ž๐ง๐ญ?

One key to successfully adopting AI into your workflow is knowing to which tasks AI is suited.
โŒ Tasks which are too complex/big will fail
โŒ Tasks which are too small don't bring you benefit

Image description

Working with a small team and heavily adopting AI to help us ship faster and build our startup, I've learned to triage the tasks I delegate to AI as follows:
โœ… Touches max 3 files
โœ… Clear success criteria
โœ… Single repository
โœ… No gotchas
โœ… Would take up to 3 hours without AI

When I go through our development backlog looking for these sorts of issues, what I find we're left with tends to include
โœ… All minor bugs
โœ… UI tweaks
โœ… Chore tasks (e.g. adding proper error handling)

In summary, what we call "low-surface context" tasks, where the path to resolution is clear and straightforward.


P.s.
๐Ÿค– I've built a script that goes through your Linear and triages your tasks according to these criteria, to add an AI-ready label. I shared it with my substack subscribers. If you're interested, comment below and I'll send you the link.
Image description

Top comments (0)