Podcast Summary: Lenny’s Podcast — Head of Claude Code: What Happens After Coding Is Solved | Boris Cherny (Anthropic)

Date: February 19, 2026
Host: Lenny Rachitsky
Guest: Boris Cherny, Head of Claude Code at Anthropic

Overview

In this episode, Lenny Rachitsky sits down with Boris Cherny, the Head of Claude Code at Anthropic. The conversation explores the astonishing transformation of software engineering since the advent of AI-powered coding tools, particularly Claude Code. The discussion covers the speed and breadth of these changes, how coding is now considered a “solved” problem, the granular tactics of building products on AI, what’s next for adjacent tech roles, AI-powered agents, principles for building and scaling AI products, and Boris’ personal perspectives, principles, and journey.

Key Discussion Points & Insights

1. The Impact of Claude Code & the AI Coding Revolution

• AI is now writing the majority of code: Boris shares that he hasn’t edited a line by hand since November, with 100% of his code produced by AI agents, and routinely shipping 10–30 pull requests per day (00:00, 16:23).
• Industry-wide shift: 4% of all public GitHub commits are now from Claude Code, with private codebases likely to be much higher (06:42). Predictions are that AI will account for 20%+ of all code commits by year-end.
• Exponential growth: The pace of change is accelerating, with daily active users doubling month-over-month (06:42).
• Changing roles: Boris envisions the software engineer role becoming obsolete, replaced by “builders” or “product managers who code” (44:44, 43:05).
• Comparison to the printing press: Coding tools are making programming accessible to everyone, which Boris likens to the democratizing effect of the printing press (32:27).

“In a year or two, it’s not going to matter. Coding is largely solved. I imagine a world where everyone is able to program. Anyone can just build software anytime.” — Boris (00:22, 39:36)

2. The Next Frontier: From Coding to Decision-Making, Agents, and Beyond

• AI agents evolve: Claude is now analyzing bug reports, feedback, and telemetry to independently propose product ideas and bug fixes — functioning more like a proactive coworker rather than a mere tool (17:54).
• Expanding beyond code: AI is increasingly automating non-coding tasks: paying parking tickets, project management, internal communications, and more (18:37).
• AI as a PM and Designer tool: Product management, design, and data science are next in line for transformation, with AI acting as “agents” that use tools, act autonomously, and blur traditional role boundaries (36:22).

“Claude is starting to come up with ideas … a little more like a coworker or something like that.” — Boris (17:54)

3. Building, Shipping, and Scaling with AI: Tactics and Principles

• Build for where the model will be, not where it is: Products should be designed to take advantage not just of current model capabilities, but anticipating major near-future leaps (63:42).
• Under-resource intentionally: Giving small teams fewer resources forces them to use and trust AI more, leading to better outcomes and innovation (24:22).
• Be generous with tokens: Especially in early stages, give engineers plenty of compute/budget to experiment; innovations often emerge when pushing boundaries (25:54).
• Latent demand: Watch what users and the AI are trying to do, even if it’s not the primary use case—then build into those emergent behaviors (47:38).
• Iterate and ship early: Release products earlier than seems comfortable, then learn from user feedback and usage—especially crucial since LLM behaviors and capabilities change so quickly (54:05).
• Don’t box the model in: Giving AI models flexibility and tool access yields better results than rigid, heavily-orchestrated systems (63:42).
• The “Bitter Lesson” (Richard Sutton): Bet on the more general model rather than heavily-tuned, narrow approaches; the more general will win over time (63:42).

4. Safety, Alignment, and Product Feedback Loops

• Three layers of safety:
  1. Mechanistic interpretability (peeking inside LLM “neurons” for alignment/safety problems)
  2. Evals (synthetic/lab testing)
  3. Real-world feedback (user-in-the-loop, shipping “research previews”)
     (54:30)
• Race to the Top with safety: Anthropic open-sources tools like code sandboxes for others to ensure safe agent use (57:15).
• Product feedback deeply integrated: Boris and team actively solicit, respond to, and act on user bug reports and feature requests, sometimes fixing bugs in minutes via AI (85:21).

“The only reason that it keeps improving is because everyone is using it, everyone is talking about it, everyone keeps giving feedback.” — Boris (75:43)

5. The Human Side: Motivation, Enjoyment, and Principles

• Enjoyment is up: 70% of engineers and PMs report enjoying their jobs more since adopting AI (“coding is delightful again”) (44:38).
• Skills atrophy and nostalgia: There is some concern about losing traditional skills, but Boris likens this to previous transitions in programming history (41:13, 32:27).
• The art vs. practicality of coding: For Boris, programming is a practical tool, but he acknowledges the intrinsic beauty for some (28:06).
• Becoming a generalist: The future belongs to those who span multiple disciplines—AI enables everyone on the team (PM, EM, designer, finance) to code and contribute (40:58).
• Personal journey: Boris’ life and perspective shaped by moving from Ukraine, living in rural Japan (making miso!), then joining Anthropic due to a sense of mission (72:30).

Notable Quotes & Memorable Moments

• On the future of coding:

  “100% of my code is written by Claude Code. I have not edited a single line by hand since November … Productivity per Engineer has increased 200%.” — Boris (00:00, 20:50)

• On product management and traditional roles:

  “I think by the end of the year everyone’s going to be a product manager and everyone codes. The title software engineer is going to start to go away. It’s just going to be replaced by builder.” — Boris (00:44, 43:05)

• On latent demand:

  “Latent demand is … users using the product in a way it wasn’t designed for—to do what they want. That helps you learn where to take the product next.” — Boris (47:38)

• On advice for AI builders:

  “Don’t try to box the model in … Always bet on the more general model.” — Boris (63:42)

• On safety:

  “We call this mechanistic interpretability. … We can monitor a neuron for deception and know when it’s activating.” — Boris (57:15)

• Personal connection:

  “I don’t think you know this, but I was born in Ukraine also.” “Oh, me too. Odessa.” — Lenny & Boris (62:27–62:34)

• Favorite motto:

  “Use common sense.” — Boris (84:14)

Timestamps for Important Segments

• AI has 'solved' coding, Boris’s workflow: 00:00, 16:23, 17:32
• Reflections on AI’s exponential impact: 06:42, 13:29, 13:57
• Claude Code’s origin and product lessons: 06:42–13:29, 15:21
• Principles for AI product building: 24:22, 25:54, 47:29, 63:42
• Safety approach & mechanistic interpretability: 54:30, 57:15
• Role changes, future of PM/design/engineering: 43:05, 44:44, 36:22
• Latent demand, examples of emergent use: 47:38–51:55
• Cowork built in <2 weeks, rapid deployment: 51:55–54:05
• Advice for listeners & AI adoption: 40:42–42:29
• Personal stories, miso making: 72:30–73:46

Pro Tips for Using Claude Code

• Use the best/current model (Opus 4.6) for maximum effectiveness.
• Start tasks in Plan Mode to co-design solutions before execution.
• Explore different interfaces (terminal, desktop, mobile, Slack, IDE) to fit your workflow.
• Leverage parallel agents (“multi-quadding”) for high productivity.
• Ask Claude Code for recommendations—it knows about its own settings and can help optimize itself. (68:51)

Final Reflections

Boris closes with a powerful vision: We are just 1% done in the AI assistive revolution. The next chapter will see everyone empowered to build, with the traditional boundaries between engineering, product, and design dissolving. The best way to seize this moment is to experiment relentlessly, embrace feedback, and anticipate the ever-accelerating pace of model capabilities.

“It just feels like this is 1% done and there’s so much more to go.” — Boris (74:18)

Find Boris:

• Twitter/X or Threads: @boriscodes — tag with bugs or feature requests!

Listen to the full episode for deeper stories, book and product recommendations, and Boris and Lenny’s shared memories.