Y Combinator Podcast — Inside Claude Code With Its Creator Boris Cherny

Date: February 17, 2026
Guest: Boris Cherny, Creator Engineer of Claude Code (Anthropic)
Host & Panel: Y Combinator team, Ben Mann, YC Partner

Episode Overview

This episode dives deep into the genesis, vision, and future of Claude Code—a revolutionary AI-powered coding assistant—directly from its creator, Boris Cherny. The discussion covers Anthropic's product philosophy, rapid iteration, latent demand, technical and cultural shifts in software engineering, and practical guidance for startup founders and engineers building with the latest AI models.

Key Themes & Discussion Points

1. Iterative & Forward-Looking Development

• Building for the Future of AI: Boris emphasizes Anthropic’s philosophy:

  "We don't build for the model of today, we build for the model six months from now." (00:00, 01:52, 43:43)
  This strategic bet on AI’s exponential progress shapes all aspects of Claude Code’s evolution.

• Constant Reinvention:

  "All of QuadCode has just been written and rewritten and rewritten and rewritten over and over and over. There is no part of QuadCode that was around six months ago." (00:09, 39:27)
  The product’s codebase and features are continually updated in response to both model advances and user feedback.

2. Origins of Claude Code: Accidents & Latent Demand

• Prototyping in Terminal: Initial development was accidental and pragmatic, choosing the terminal for simplicity and speed.

  "There was no pressure because we didn't even know what we wanted to build. The team was just in Explore mode." (04:43)

• Spotting Latent Demand: Real value emerges from observing how people naturally use tools—focusing on what they already try to do.

  "Probably the single, for me biggest principle in product is latent demand. And just every bit of this product is built through latent demand after the initial CLI." (07:57, 28:47)

3. User-Centric Iteration

• Dogfooding & Organic Growth: The earliest adopters were Anthropic engineers—no mandates, just internal word-of-mouth and usage.

  "I just posted about it and they'd just been like, telling each other about it. Honestly, it was just accidental." (06:37)

• Design By Feedback: New features (“plan mode”, verbosity controls) arrive via listening to real usage and conversation:

  "Literally this was like Sunday night at 10pm...I just wrote this thing in like 30 minutes and then shipped it that night." (25:45)

4. Changing Landscape of Software Engineering

• Productivity Leap:

  "Productivity per engineer grew something like 70%...since QuadCode came out, productivity per engineer at Anthropic has grown 150%." (40:30)
  "I think the title software engineer is going to go away...Everyone on our team codes—engineers, PMs, designers, EMs, finance." (43:37)

• Generalists and Multi-Disciplinary Teams: The most successful engineers are either extreme specialists or “hypergeneralists.” (18:53)

5. Founder and Hiring Advice in the AI Era

• Beginner’s Mindset & Humility:

  "Engineers, as a discipline, we've learned to have very strong opinions...but it actually turns out a lot of this stuff just isn't relevant anymore." (16:05)

• First Principles & Willingness to be Wrong:

  "I sometimes ask about what's an example of when you're wrong...For me personally, I'm wrong probably half the time." (16:42)

• Hiring via Agent Interaction: Evaluate candidates by how well they interact with AI agents—not just through traditional interviews.

  "You can upload a transcript of you coding a feature with Claude code...you can figure out how someone thinks..." (17:32)

6. Claude Code's Architecture and Usage Patterns

• Terminal as a Form Factor:
  • Minimal UI, highly iterative:

    "We felt there is no UI we could build that would still be relevant in six months because the model was improving so quickly." (08:35)

  • Accidental longevity:

    "If you asked me this a year ago, I would have said the terminal has a three month lifespan..." (30:17)

• Plan Mode and Beyond:
  • Plan mode arose due to user workflows, but is expected to become obsolete as models improve.

    "Plan mode probably has a limited lifespan..." (25:07)
    "Claude code can now enter plan mode by itself..." (25:19)

7. Collaboration & Swarms of Agents

• Agent Topologies:
  • The product vision includes collaborative “swarms” of agents working in parallel, especially for complex tasks.
  • Example: The entire Plugins feature was built by a swarm of agents with minimal human intervention—agents self-organized, wrote tasks, and executed.

    "The Plugins feature was entirely built by a swarm over a weekend..." (21:59, 23:02)

8. Broader Impact and Future Vision

• Trajectory & Exponential Progress:

  "Really all it is is you just trace the exponential...Coding will be generally solved for everyone." (43:37)

• The Bitter Lesson Philosophy:
  • Emphasizes the inevitability of more general learning systems winning over hand-crafted scaffolding/code.

    "Never bet against the model...assume that whatever the scaffolding is, it's just tech debt." (43:43, 37:43)

• Mission-Driven Focus & Safety:
  • Anthropic’s drive is not just about speed, but about building safe systems—AI safety is discussed constantly within the organization. (41:45)

Memorable Quotes & Moments

• On Model-First Mindset:

  "Build for the model of six months from now." — Boris Cherny (00:00, 43:43)

• On Accidental Product-Market Fit:

  "It just wants to use tools. That's how it wants." — Boris Cherny (05:24)

• On Early Usefulness:

  "Robert...just had quad code on his computer and he was using it to code. I was like, what? What are you doing? Like, this thing isn't ready. It's just a prototype. But, yeah, it was already useful in that form factor." — Boris Cherny (06:24)

• On User Feedback:

  "I remember early on, I tried to get rid of BASH output...everyone just revolted. I want to see my dash." — Boris Cherny (12:05)

• On Plan Mode:

  "There's no big secret to it. All it does is it adds one sentence to the prompt that's like, please don't code." — Boris Cherny (25:26)

• On Productivity Leap:

  "The team doubled in size last year, but productivity per engineer grew something like 70%...since Quadcode came out, productivity per engineer at Anthropic has grown 150%." — Boris Cherny (40:30)

• On the Future of Engineering:

  "I think we're going to start to see the title software engineer go away." — Boris Cherny (43:37)

• On the Bitter Lesson:

  "The idea is the more general model will always beat the more specific model." — Boris Cherny (37:43)

Timestamps of Key Segments

| Time | Segment | |------------|-------------------------------------------------------------------------------------------------------| | 00:00 | Boris explains Anthropic’s model-first, six-months-ahead product philosophy | | 01:19 | The inception of Claude Code; “three months, no vacation” | | 04:23 | Early days: prototyping with the terminal, Bash, and tool use in Anthropic Labs | | 05:24 | First “AGI moment”: model scripting AppleScript | | 06:07 | Internal adoption: “What? You’re using this?” | | 07:17 | Early use cases: automation, Git, unit tests, Markdown/Claude MD emerges | | 09:01 | Boris shares his minimal Claude MD, importance of lean user instructions | | 12:05 | Internal revolt against hiding Bash output—listening to users | | 16:05 | Advice for founders: staying humble, beginner’s mindset, and learning to be wrong | | 17:32 | Discussing hiring via agent/coding transcripts | | 18:53 | Two archetypes of effective engineers: hyperspecialists and hypergeneralists | | 21:59 | Vision for Claude Teams and agent topologies | | 23:02 | Plugins feature “built by a swarm”—process and outcomes | | 25:07 | Discussion of Plan Mode’s inevitable obsolescence | | 28:47 | Latent demand & origins of Plan Mode | | 30:17 | Terminal’s longevity, evolving interface experiments | | 31:11 | DevTool founder advice: serving both engineers and agent desires | | 34:43 | Product parallels: TypeScript’s practical design, build for how people really work | | 37:43 | The Bitter Lesson: betting on general models over scaffolding, tech debt mindset | | 39:27 | Codebase churn: “rewritten and rewritten,” rapid obsolescence | | 40:30 | Internal productivity metrics—1000x improvements, industry transformation | | 41:45 | Boris’s motivation: AI transition, safety, and mission-driven culture at Anthropic | | 43:37 | The future: end of "software engineer" title, generalist teams, and dangers as AI scales | | 46:21 | Claude Code’s adoption surge, NASA/Mars Rover, prevalence in startups | | 47:47 | The rise of Cowork: non-technical use cases and GUI interfaces | | 49:32 | Closing: host’s thanks, Boris’s invitation to send bugs |

Practical Takeaways for Founders & Builders

• Build for the AI model that’s coming, not the one that exists today.
• Search for “latent demand” by studying user behaviors; don’t force new workflows.
• Ship minimal UI—function matters most when platform capabilities are evolving rapidly.
• Invest in scientific mindset, humility, and rapid iteration—old expertise may not transfer.
• Hire and partner with people who “think weird”—generalists and outliers can thrive now.
• Assume all scaffolding and custom code is temporary; prefer waiting for model progress over hand-coded patches.
• Be vigilant about AI safety: adopt a mission-driven culture as capabilities rapidly increase.

Notable Anecdotes

• NASA uses Claude Code to plot the Mars Rover course (46:21).
• The original plan mode feature was coded and shipped in 30 minutes based on user feedback (25:45).
• Anthropic has a framed copy of Rich Sutton's The Bitter Lesson in the Claude Code team area, underscoring their philosophy (37:43).
• “Plan mode” and other features are expected to become obsolete as models autonomously improve (25:07, 26:47).
• Productivity at Anthropic has increased 1000x compared to Google in its peak engineering years (40:00).

End of summary.