Hey Product Hunt. I built ManualMode because coding agents made me faster, but I noticed they also removed a lot of the small work that used to teach me a codebase: tracing the odd failure, reading the test twice, and catching a bad assumption in a clean-looking patch. ManualMode keeps one useful piece of that work in the loop. Your coding agent can reserve a bounded task from your own repository for you to complete manually. When no project task fits, the Gym gives you a short browser rep in code reading, debugging, test design, async reasoning, or AI review. You can start free with three Gym reps and one Project rep. I would especially like feedback from engineers who use Codex, Claude Code, Cursor, or similar agents every day: does the reserved task feel useful, or does it feel like artificial friction?

ManualMode

Coding fitness for engineers who ship with AI agents

Coding fitness for engineers who ship with AI agents