Boris Cherny on Trying to Get Claude Code to Rewrite the Claude App
AI Summary
Daring Fireball links to Boris Cherny’s account of trying to use Claude Code to rewrite the Claude app. The associated Hacker News thread had 69 points and 14 comments at the captured snapshot. The supplied metadata does not describe the rewrite’s scope, completion level, architecture, code quality, human intervention, or measured results. This is therefore best treated as a timely practitioner signal about agentic coding ambitions, not as validated evidence that an AI coding agent can autonomously rebuild a production application.
Why it's worth reading
AI coding agents are being pushed from local edits toward application-scale work, making this attempt timely, although the original account is needed to judge autonomy and actual results.
Deep Read
1. What happened
Original facts: Daring Fireball published a linked item whose title says Boris Cherny tried to get Claude Code to rewrite the Claude app. The associated Hacker News snapshot shows 69 points and 14 comments.
2. Core technology
Original facts: The supplied material identifies Claude Code, the Claude app, and an attempted “rewrite.”
Unknown: It does not specify the repository, language or framework, agent permissions, prompting strategy, test workflow, context management, or level of human participation. The technical approach therefore cannot be reconstructed reliably.
3. Key evidence and numbers
- Published: 2026-08-03T00:22:55.000Z
- Hacker News snapshot: 69 points and 14 comments
- Named products: Claude Code and the Claude app
- Not provided: lines of code, duration, cost, test pass rate, defect rate, performance comparison, or production status
4. Why it matters
Analysis: An application-scale rewrite is a more demanding test of coding agents than completion, isolated bug fixes, or small refactors. It can expose limits in long-context reasoning, cross-file consistency, testing loops, and architectural judgment. However, the word “trying” does not establish that the rewrite succeeded.
5. Practical impact
Analysis: Engineering teams can use cases like this to structure bounded agent experiments: select a module, define executable acceptance tests, and record human intervention, cost, and regressions. The available summary is not sufficient evidence for migrating production code or reducing human review.
6. Limitations and uncertainty
The input consists of a Daring Fireball link page and Hacker News metadata, not Cherny’s complete original account. The rewrite’s scope, outcome, production use, and degree of Claude Code autonomy remain unverified. Hacker News points and comment counts can also change over time.