Claude Code creator Boris Cherny shares five practical Agent principles in a Stanford class
Claude Code creator Boris Cherny shared five practical principles in week four of Stanford's CS146S course. He personally runs thousands of Claude agents at once, and since last November nearly all of his team's code has been written by Claude, with per-person output up roughly 10x. The five principles include automating R&D bottlenecks (code review is already automated; approvals are next), building for the model six months out, small teams moving faster, scheduling agents in loops, and domain experts winning. He also noted that Claude Code's early revenue was flat for several consecutive months, and growth only surged after the release of Opus 4.
Why read this
AI score 67
The Claude Code creator's approach of automating R&D bottlenecks and designing ahead for future model capabilities offers a transferable engineering playbook for teams.
Topics
View original postI asked @bcherny how many Claude agents he's running right now as he sat next to me. "A few thousand." Boris (creator of Claude Code) came back to my Stanford course The Modern Software Developer last Thursday. His main takeaways: - Automate your bottleneck: Claude has written 100% of his team's code since November and output is up ~10x per engineer. So much code shipped that people stopped reading it carefully and started skimming, so they automated code review. Approvals are next. The team identified the main things impeding their development velocity and built systems to unblock themselves. - Build for the model 6 months out: early Claude Code revenue was flat for months because the model wasn't smart enough yet. Then Opus 4 came out and growth became astronomic. His method is to read your agent transcripts and see where the model fails. That's where it will improve in the next model release. - Small teams go faster: for months Claude Code was basically just him. With AI, one or two people can outship a team 100 times their size. Constraints breed creativity. - Think in loops: his grandpa programmed punch cards in the Soviet Union. Boris writes loops that run for days or weeks and spawn many agents. He expects that typing code by hand is heading where punch cards went. - Domain-specific builders are the future: he expects the best accounting software of 2027 to be built by an accountant. When code is cheap, knowing the problem deeply is the only alpha In the 50 years after the printing press, Europe produced more literature than in the previous 1,000. He thinks coding is about to undergo a similar explosion of creation.
View original
I asked @bcherny how many Claude agents he's running right now as he sat next to me. "A few thousand." Boris (creator of Claude Code) came back to my Stanford course 𝗧𝗵𝗲 𝗠𝗼𝗱𝗲𝗿𝗻 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿 last Thursday. His main takeaways: - 𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗲 𝘆𝗼𝘂𝗿 𝗯𝗼𝘁𝘁𝗹𝗲𝗻𝗲𝗰𝗸: Claude has written 100% of his team's code since November and output is up ~10x per engineer. So much code shipped that people stopped reading it carefully and started skimming, so they automated code review. Approvals are next. The team identified the main things impeding their development velocity and built systems to unblock themselves. - 𝗕𝘂𝗶𝗹𝗱 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗺𝗼𝗱𝗲𝗹 𝟲 𝗺𝗼𝗻𝘁𝗵𝘀 𝗼𝘂𝘁: early Claude Code revenue was flat for months because the model wasn't smart enough yet. Then Opus 4 came out and growth became astronomic. His method is to read your agent transcripts and see where the model fails. That's where it will improve in the next model release. - 𝗦𝗺𝗮𝗹𝗹 𝘁𝗲𝗮𝗺𝘀 𝗴𝗼 𝗳𝗮𝘀𝘁𝗲𝗿: for months Claude Code was basically just him. With AI, one or two people can outship a team 100 times their size. Constraints breed creativity. - 𝗧𝗵𝗶𝗻𝗸 𝗶𝗻 𝗹𝗼𝗼𝗽𝘀: his grandpa programmed punch cards in the Soviet Union. Boris writes loops that run for days or weeks and spawn many agents. He expects that typing code by hand is heading where punch cards went. - 𝗗𝗼𝗺𝗮𝗶𝗻-𝘀𝗽𝗲𝗰𝗶𝗳𝗶𝗰 𝗯𝘂𝗶𝗹𝗱𝗲𝗿𝘀 𝗮𝗿𝗲 𝘁𝗵𝗲 𝗳𝘂𝘁𝘂𝗿𝗲: he expects the best accounting software of 2027 to be built by an accountant. When code is cheap, knowing the problem deeply is the only alpha In the 50 years after the printing press, Europe produced more literature than in the previous 1,000. He thinks coding is about to undergo a similar explosion of creation.
Body · AI translation
@stanfordnlp Professor @mihail_eric's new fall course "CS146S: The Modern Software Developer", course intro: https://t.co/AhFJW6xl8V
In week four of the course, the author of Claude Code, Boris, was invited to share. The core of his talk was not "use AI more to write code," but to systematically automate the R&D bottlenecks, and to design ahead for the stronger models six months from now. He himself runs "thousands" of Claude agents simultaneously; since last November, his team's code has been almost entirely written by Claude, with per-person output up about 10x. https://t.co/9iItZYGIze
# Five key points
1. Automate your bottlenecks. Once too much code is being written, people start just glancing at it and no longer reading it closely, so they automated code review; the next step is approval. The approach is to first identify the steps that slow down delivery, then build systems for those steps, rather than vaguely "AI-ifying everything."
2. Build for the model six months from now. Claude Code's early revenue was nearly flat for several consecutive months because the models at the time weren't strong enough; growth only took off after Opus 4 came out. His method: read agent conversation logs to see where the model fails now; that is often exactly the capability the next version of the model will fill in.
3. Small teams move faster. For several months Claude Code was basically just him. He believes that with AI, one or two people can outperform a team a hundred times larger; constraints instead spur creativity.
4. Think in "loops." His grandfather programmed with punch cards in the Soviet Union; he himself writes loops that can run for days or even weeks and continuously spin up large numbers of agents. He expects hand-typing code to gradually become an obsolete skill, like punch cards.
5. People who understand the domain will win. He predicts that in 2027 the best accounting software will be built by accountants. When writing code becomes cheap, the real barrier is a deep understanding of the problem itself.