How does an AI coding agent work?
Watch “Add dark mode” become a change with evidence.
- CHANGED
- —
- VERIFIED
- Awaiting evidence
- BLOCKERS
- Checks incomplete
One request. Follow the work, including the checks.
- ○ Context
- ○ Diff
- ○ Tests
- ○ Build
- ○ UI
What’s real, what’s simplified
Real: A coding agent can read selected files, change code and use check results as feedback when it has permission.
Simplified: All files, changes and checks are curated visuals, not executed commands. The mobile example repairs an oversized theme button. Human approval is simulated; no AI, repository access or real publishing occurs.
Can vary: Different agents have different tools, checks and approval rules. Passing these example checks does not prove every possible behavior. This is a local visual, not a real agent or private reasoning transcript.
Sources
This is a curated browser-only simulation: no AI calls, uploads, repository access or real commands. Concise process summaries, not private reasoning transcripts.