Terminal-first AI — the work that should not go to an editor
The division is simple once you see it: work driven by command output belongs where the commands run.
AI Tools Hub Editorial · published · updated
Terminal-shaped work
- Dependency updates and the build failures that follow
- Finding anomalies in logs
- Writing a migration script while testing it
- Chasing a CI failure
- Applying the same change across many files
What these share is a loop: run a command, read the output, decide the next command. Carrying that loop into an editor chat means copying output back and forth, which costs more than the AI saves.
Editor-shaped work
- Writing a new feature
- Refactoring
- Fixing type errors
- Writing tests
Running work from elsewhere
Warp accepts @Warp mentions from Slack, Linear and GitHub. On an incident you can start the investigation immediately and read the result after finishing what you were doing. That is not automation; it moves the start time earlier, which shortens elapsed time.
Decide what the agent may run
Before letting an agent execute commands, decide the boundary. Deletions, production connections and credential reads should require confirmation. If you want the execution itself isolated, that is a sandbox question — E2B rather than your laptop.
FAQ
Do I have to switch terminals?
Warp ships a CLI, so you can use the agent inside your existing terminal first. Lower friction than a full migration.