AgentDiff Review Assistant
A structured review view for codebases heavily modified by AI agents
- Problem
- AI agents can now make sweeping changes across a repo, leaving developers feeling out of control about what changed and whether to accept it. One Ask HN asks how to review code in the new AI world, while another notes that PR review was already hard and now that AI writes and reviews code, they get even less feedback than before.
- Solution
- Aggregate agent-generated diffs into semantic units, generate collapsible change summaries, flag risky edits, and let users accept or reject block by block. The core is helping humans grasp what an agent did in minutes instead of reading raw diffs.
- Users
- Individual developers and small teams using Claude Code, Codex, or Cursor who need to review agent output daily.
- MVP
- Read git diffs and agent session logs; cluster changes by file and function; generate a human-readable change narrative; provide block-level accept/reject; record review decisions for later recall.
- Revenue
- Free viewing tier; subscription unlocks team rules, review history, and export; per-seat monthly pricing.
- Why now
- Agents write code much faster than humans can review it, making review the new bottleneck; an indie dev can own the review experience layer without touching models.