Critic
Critic helps developers understand and review the code their AI agents write, with collaborative code review and agent question tracking across sessions.
At a Glance
About Critic
Critic is a code review and agent oversight tool built by Feyn (Chonkie, Inc.), a Y Combinator-backed company based in San Francisco. It surfaces the changes AI coding agents make, lets human reviewers ask questions, and keeps those questions alive even when devices disconnect. The product is accessible via the web and is currently in an early preview stage, as indicated by its queue position on the homepage.
What It Is
Critic sits between a developer's AI coding agents (such as Codex or Claude) and the main codebase, providing a structured review layer for agent-generated code changes. Rather than treating agent output as a black box, Critic presents each set of changes with a narrative explaining what changed and why, and allows reviewers to ask targeted questions that the agent or author must answer before the code merges.
How the Review Workflow Works
Each change set in Critic includes a "Change Narrative" that describes the before and after state of the code. Reviewers — both human and agent — can post questions tied to specific changes. The system tracks how many messages each question thread has generated and surfaces open questions prominently. The homepage example shows a real workflow: a branch (codex/durable-agent-questions → main) with three commits, three changed files, and multiple open reviewer questions from named team members.
Durable Agent Questions Across Disconnects
A notable technical feature visible on the homepage is the leased handoff system for agent questions. When a reviewer asks a question from a change page, the question waits for the author's coding agent rather than assuming the author's machine is online. A short lease (45 seconds) lets any connected author device claim the work atomically; if a device loses its network connection mid-answer and the lease expires, another device can pick up the question without losing the thread. The homepage shows the implementation detail: a convex/agentQuestions.ts file that "turns a best-effort queue into a leased handoff."
Anchor Movement and File Change Display
Critic also handles the problem of code anchors shifting as files change. The packages/protocol/src/project-anchor.ts file shown on the homepage implements conservative anchor movement: exact blob matches win, while rename and surrounding-context candidates are scored, with close matches surfaced as ambiguous rather than silently attached to the wrong line. This prevents review comments from drifting to incorrect locations as agents continue editing.
Current Status
Critic is in early preview — the homepage shows a queue position (#1843/3), suggesting a waitlist or limited access model. The product is built by Feyn (operating as Chonkie, Inc.) and is backed by Y Combinator. Sign-in is available via GitHub. The linked GitHub repository used in the homepage demo (kelseyhightower/nocode) is a well-known satirical open-source project, not Critic's own source code.
Community Discussions
Be the first to start a conversation about Critic
Share your experience with Critic, ask questions, or help others learn from your insights.
Pricing
Preview
Early preview access to Critic via waitlist, sign in with GitHub.
- GitHub sign-in
- Agent code review
- Change narrative
- Reviewer question threads
- Local preview
Capabilities
Key Features
- AI agent code review
- Change narrative generation
- Reviewer question threads
- Durable agent questions across device disconnects
- Leased handoff for agent question ownership
- Conservative code anchor movement
- GitHub sign-in
- Multi-reviewer support
- Branch-to-main change tracking
- Local preview support
