Boris Cherny, the Anthropic engineer behind Claude Code, has a simple mental model for getting useful output from AI: treat it like a colleague sitting across from you. In a Business Insider interview, Cherny laid out three practical prompting tips that follow directly from that mindset, offering a concrete window into how one of the technology's own builders actually uses it day to day.
The Coworker Framework
Cherny's core argument is that most people prompt AI the way they use a search engine: short queries, minimal context, and an expectation that the system will fill in the blanks. That approach tends to produce generic results. When you talk to a coworker about a problem, you give them background, you explain what you've already tried, and you tell them what a good answer looks like. Cherny says prompting Claude works the same way. For more on how Cherny thinks about his work with AI agents, his approach to managing tens of thousands of AI agents daily gives useful context on the scale at which he operates.
Key Facts
- Boris Cherny is the creator of Claude Code, Anthropic's agentic coding tool.
- His three tips center on context, iteration, and explicit goal-setting.
- Cherny says he has not written code himself in several months, relying instead on AI-assisted workflows.
- The tips apply broadly across Claude models, not just to coding tasks.
The first tip is to front-load context. Rather than asking a bare question, Cherny recommends opening with the situation: what the project is, what constraints exist, and what has already been tried. The second tip is to state the desired output format explicitly. If you want a bulleted list, say so. If you want a rough draft rather than a polished document, say that too. Vague requests tend to produce vague answers. The third tip is to iterate rather than restart. If the first response misses the mark, push back directly and explain why, the same way you would give feedback to a human colleague rather than walking away and starting a new conversation.
"I talk to Claude the way I'd talk to a really smart new hire who doesn't know our codebase yet. I wouldn't just say 'fix this bug.' I'd explain what the bug is, what I think is causing it, and what a good fix would look like."Boris Cherny, Business Insider
Practical Implications for Everyday Users
What makes Cherny's advice notable is that it comes from someone who has spent significant time on the engineering side of these tools. He is not describing theoretical best practices. He is describing what he does himself. As reported earlier this year, Cherny has gone months without writing a line of code himself, delegating that work entirely to AI-assisted workflows. His prompting habits are therefore load-bearing in his actual job, not just an interesting experiment.
The tips also connect to a broader conversation in the developer community about how to measure and improve AI-assisted work. Cherny has previously argued for rethinking how teams evaluate AI productivity, suggesting that token consumption is a poor proxy for value created. His case for a better metric than token burn points in a similar direction: the quality of the human-AI interaction matters more than the raw volume of output. These prompting habits are part of the same thinking.
For teams debating how much to invest in structured prompting practices versus just experimenting freely, Cherny has also weighed in on that balance. He has argued that an ROI focus is reasonable, but that leaving room for open-ended experimentation is equally important for teams that want to stay ahead of what the tools can actually do.
The three tips Cherny outlines are not technically complex. They do not require special tools or prompt engineering expertise. They require a shift in how people think about what they are doing when they type into a chat window. That shift, from query-sender to collaborator, may be the more difficult adjustment, but it is also the one Cherny thinks makes the biggest practical difference.
