Episode Summary
Executive Summary: Boris Cherny explains how Claude Code emerged from experimenting for the next model, not the current one, and how user feedback shaped its terminal-first workflow, plan mode, and lightweight custom instructions. He argues that product velocity now comes from building with models, not around them, and that coding is rapidly becoming agentic, collaborative, and increasingly automated across the org.
Main Topics: Building for the next model, not today’s model (Priority: 5/5): Cherny’s central philosophy is to design for the frontier model expected in six months, because present-day scaffolding often becomes obsolete quickly as model capabilities improve. The accidental origin and terminal-first form factor of Claude Code (Priority: 5/5): Claude Code began as a simple terminal app to test Anthropic’s API and tool use; the CLI remained because it matched latent demand and was fast to prototype, despite expectations it would evolve into an IDE-like product. Latent demand and iterative product development (Priority: 5/5): The team repeatedly shipped small changes based on observed user behavior and GitHub feedback, adding features like plan mode, verbosity controls, and QuadMD only when users demonstrated the need. Agentic workflows, sub-agents, and team collaboration (Priority: 4/5): Cherny describes how Claude Code now spawns sub-agents, uses fresh context windows, and coordinates work through tools like Asana, Slack, and GitHub to parallelize debugging and product development. Hiring, engineering culture, and evaluating human judgment in the AI era (Priority: 4/5): He emphasizes humility, first-principles thinking, and the ability to learn from mistakes as key traits, and suggests Claude Code transcripts may become a useful signal for hiring and evaluation. Productivity gains and the shift in software work (Priority: 5/5): Anthropic reports major productivity gains from internal Claude Code usage, and Cherny predicts coding becomes broadly solved, with engineers increasingly acting as generalist builders rather than pure coders. Safety, mission, and future risk (Priority: 4/5): Cherny frames Anthropic’s mission-driven culture as central, noting that scaling model capability carries both enormous upside and serious risks like recursive self-improvement and misuse.
Key Arguments: Build for the model six months ahead because model improvements will quickly erase bespoke scaffolding gains. Latent demand is the most important product principle: make easier what users are already trying to do, not ask them to do something new. Claude Code stayed in the terminal because it was the cheapest, fastest prototype and users proved the form factor worked. Many features came from user behavior, such as markdown instructions becoming QuadMD and plan mode emerging from users asking Claude to think before coding. As models improve, less explicit prompting and less scaffolding will be needed; plan mode may eventually become unnecessary. The best engineering talent now combines humility, scientific thinking, and first-principles reasoning rather than fixed strong opinions. Agentic coding will broaden beyond engineers; designers, finance, and sales increasingly use these tools, and coding itself is becoming more universal. The future of software work is generalist: specs, user interaction, testing, and coordination become as important as writing code. Anthropic’s mission and safety focus matter because more capable models increase both opportunity and catastrophic risk. Transcript-based agent usage may become a strong hiring signal because it reveals systems thinking, debugging habits, and how people correct agents.
Data Points: Prototype timeline: 2 days - Cherny said he gave the first prototype to his team for dogfooding just two days after building it. Early initial focus period: September 2024 - He said he worked through nights and weekends for about three straight months starting in September 2024. Internal launch review date: December 2024 - He referenced Dario asking about internal usage during a launch review around December 2024. Early code generation share: about 10% of his code - In February, Claude Code was only writing a small fraction of his code. Current internal code written by Claude: 70% to 90% - He said Anthropic teams vary, but internal code generation by Claude Code ranges in this band. Personal current code written by Claude: 100% - Cherny said that since Opus 4.5 he has not edited code by hand. PR volume: 20 PRs a day - He claimed he lands around 20 pull requests daily. Productivity increase at Anthropic since Claude Code: 150% - He said productivity per engineer has grown by this amount since Claude Code came out. Team growth last year: doubled in size - He contrasted team growth with productivity gains. Estimated productivity growth last year: 70% - He said productivity per engineer rose about 70% as the team doubled in size. Coding automation prediction: 90% - He cited Dario’s prediction that 90% of Anthropic code would be written by Claude. External usage stat: 70% of startups - He mentioned a Mercury stat that 70% of startups choose Claude as their model of choice. Public commit share: 4% - He cited a semi-analysis stat that 4% of all public commits are made by Claude Code. Model release trend: every few months - He argued new models arrive frequently enough that scaffolding can be wiped out quickly. Internal product rewrite cadence: every couple weeks - He said they unship tools and add new tools on this cadence.
Pivotal Quotes: "We don't build for the model of today, we build for the model six months from now." — Boris Cherny: His core product philosophy for building Claude Code and advising founders. "Latent demand is probably the single biggest principle in product." — Boris Cherny: He explains why Claude Code features emerged from observing what users already wanted to do. "There's no part of quad code that was around six months ago." — Boris Cherny: He describes how quickly the product and scaffolding are rewritten as models improve.
Implications: AI devtools will reward speed, iteration, and user-observed demand over static architecture. As models improve, coding, debugging, and even product coordination become increasingly agent-driven, pushing software teams toward generalist, system-oriented workflows.
From the Transcript
At Anthropic, the way that we thought about it is we don't build for the model of today, we build for the model six months from now. That's actually like still my advice to founders that are building on LLMs. Just try to think about like what is that frontier where the model is not very good at today? Because it's gonna get good at it. All of quad code has just been written and rewritten over and over. There is no part of quad code that was around six months ago. You try a thing, you give it to users, you talk to users, you learn, and then eventually you might end up at a good idea, sometimes you don't. Are you also in the back of your mind thinking that maybe like in six months you Need to prompt that explicitly, but the model will just be good enough to figure out on its own. Maybe in a month. No more need for a plan mode in a month. Oh my god.
How often do you rewrite the code base of clock code? Is this every six months with this first case? Is there scaffolding that you've deleted because you don't need it anymore? Because the model just improved. Oh, so much. Yeah. Like all of quad code has just been written and rewritten over and over. We unship tools every couple of weeks. We add new tools every couple weeks. There's no part of quad code that was around six months ago. It's just constantly rewritten. But you say most of the code base for our current clock code is only, say, 80% of it is. Only less than a couple months old. Yeah, definitely. It might even be like less than, yeah, maybe like a couple months. That feels about right. So it's like the life cycle of code now. That's another alpha is expecting it to be the shelf life to be just a couple months for the best founders. Did you see Steve Yeggy's post about how awesome working at Anthropic is? And I think there's a line in there that says that an Anthropic engineer currently averages 1000x more productivity than a Google engineer at Google's peak.
About Y Combinator Startup Podcast
We help founders make something people want. The Y Combinator Podcast is where builders talk about building. From the earliest days of an idea to scaling a company that changes the world, YC partners and founders share real stories, lessons, and tactics from the frontlines.