Boris Cherny on Trying to Get Claude Code to Rewrite the Claude App
daringfireball.net
August 3, 2026
4 min read
🔥🔥🔥🔥🔥
45/100
Summary
Boris Cherny, head of Claude Code at Anthropic, emphasized the importance of challenging Claude with difficult tasks rather than focusing solely on prompt engineering. He discussed strategies for directing Claude to tackle complex assignments during an interview at Y Combinator’s Startup School 2026.
Key Takeaways
- Boris Cherny, head of Claude Code at Anthropic, emphasized the importance of task verification when directing Claude to perform complex tasks.
- Cherny initiated an experiment where Claude was tasked with rewriting the Electron app in Swift, which has been running for over two weeks.
- The current Claude Mac app has improved significantly in speed and reliability over the past six months, according to Cherny.
- Critics argue that simply rewriting the existing poorly designed app in a native framework will not address fundamental design issues.
Community Sentiment
Positives
- Boris raises an important point about verification — it's the single most crucial aspect that often gets overlooked in AI-driven app development.
- The adaptation of model-based systems engineering to improve verification processes is an innovative approach that could enhance reliability in AI applications.
Concerns
- The results from Claude models are flaky, with some users reporting better performance from older models like Meta Llama 4 — that’s a troubling sign for Claude's current capabilities.
- Relying on LLMs for parsing and UI tasks is misguided; they're terrible at actual parsing and need human input to be effective.
- Two weeks of token usage could cost tens of thousands, suggesting that hiring a skilled developer might yield better results than relying on LLMs for complex tasks.
