Week 1
42%
success rate · 18 tasks
Agents relearn project conventions each session.
Sample timeline
Learning loop — 3-week sample
Illustrative data from a fictional repo. Shows how decision → outcome feedback compounds over time.
Week 2
61%
success rate · 24 tasks
memory_search + experience events reduce repeat mistakes.
Week 3
78%
success rate · 27 tasks
Handoffs carry file scope and decisions; fewer duplicate branches.
Recent outcomes (sample)
- Fix billing webhook idempotencycodex · week 3 · success
- Add pricing outcome footnoteclaude · week 3 · success
- Retry swarm claim race testcursor · week 2 · partial
These are sample outcomes. Run 0dai on your own project to see real ones.