The PodKnowledge|← All Episodes
Based on Lenny's Podcast data
Lenny's Knowledge Sketch

This Is the
World Now

Boris Cherny
Head of Claude Code, Anthropic (fmr. Instagram, Meta)
FEB 19 2026
The Premise

Every Line, Written
by Claude

SPEC / IDEA5 AGENTS10-30 PRs/DAY
"100% of my code is written by Claude Code. I have not edited a single line by hand since November."
  • Ships 10, 20, 30 pull requests a day, five agents running at any moment
  • Started at 20% of his code (Feb), 30% (May), crossed 100% in November
  • Enjoys coding more now, "because I don't have to deal with all the minutia"
  • Still reviews the code; a human check stays in the loop for anything that runs in production
Framework

The Curve, Named

FEB 20%MAY 30%NOV 100%BORIS'S OWN CODE, WRITTEN BY CLAUDE
4%
of GitHub commits are Claude Code today
1/5
predicted by year end (Semi-Analysis)
2×
Claude Code DAU, past month
  • Anthropic thinks in exponentials; three co-founders wrote the scaling-laws paper
  • Boris's May 2025 prediction: "You may not need an IDE to code by year end." The room audibly gasped
  • Model gains transfer sideways: teach coding, get better tool use, get better computer use
  • Any scaffolding you build gets 10-20% gains, then the next model wipes them out
The exponential ruleIf the line is exponential, trace it forward. It's already obvious you cross 100% by year end. Intuition just refuses to accept that.
Inside Anthropic

What the Data Actually Shows

PRs / engineer+200%Eng team size~4× in a yearPRs Claude reviews100%Engineers loving AI70%PMs loving AI70%Designers loving AI55%
The token advice

Boris's counsel to CTOs: don't cost-cut, don't optimize. Give engineers as many tokens as possible. Some Anthropic engineers spend hundreds of thousands of dollars on tokens a month.

Why safety, not speed

Boris left Anthropic for Cursor, then went back two weeks later. The pull was the mission. Everyone at Anthropic answers "safety" when you ask why they're there.

Playbook

Build for the World That's Coming

  • Give engineers as many tokens as they can use; don't try to optimize cost early
  • Build for where the models will be, not where they are; scaffolding today gets eaten next release
  • Give people psychological safety to fail; 80% of ideas are bad, that's fine
  • Hold them accountable to cut losses fast on the bad ones instead of investing more
The Claude Code originStarted as a personal hack. Boris spent one month building weird prototypes at Anthropic before anything shipped. Ben Mann nudged him to add a DAU chart. It went vertical.
Contrarian

The Uncomfortable Predictions

✗Should I learn to code?INSTEAD →✓ In a year or two it won't matter. Coding is virtually solved.
✗Software engineer is a titleINSTEAD →✓ By year end it starts to go away. Replaced by "builder." Everyone is a PM and everyone codes.
✗AI can only follow instructionsINSTEAD →✓ Claude is starting to come up with ideas. Reading bug reports, telemetry, and proposing what to ship next.
✗This is a niche shift for engineersINSTEAD →✓ Closest historical analog is the printing press. Literacy went from sub-1% to 70% globally, and everything downstream changed.
Based on Boris Cherny's episode on Lenny's Podcast. All ideas on this page are from the episode.Watch on YouTubeFollow @bcherny on X
GO DEEPER IN THE EPISODE
03:45 Why Boris briefly left Anthropic for Cursor (and what brought him back)05:35 One year of Claude Code08:41 The origin story of Claude Code13:29 How fast AI is transforming software development15:01 The importance of experimentation in AI innovation16:17 Boris’s current coding workflow (100% AI-written)17:32 The next frontier22:24 The downside of rapid innovation24:02 Principles for the Claude Code team26:48 Why you should give engineers unlimited tokens27:55 Will coding skills still matter in the future?32:15 The printing press analogy for AI’s impact36:01 Which roles will AI transform next?40:41 Tips for succeeding in the AI era44:37 Poll: Which roles are enjoying their jobs more with AI46:32 The principle of latent demand in product development51:53 How Cowork was built in just 10 days54:04 The three layers of AI safety at Anthropic59:35 Anxiety when AI agents aren’t working01:02:25 Boris’s Ukrainian roots01:03:21 Advice for building AI products01:08:38 Pro tips for using Claude Code effectively01:11:16 Thoughts on Codex01:12:13 Boris’s post-AGI plans01:14:02 Lightning round and final thoughts
Every sketch here began as a transcript.

Send us yours and we’ll draw it. $1 a transcript.

Sketch mine →
𝕏︎ X / Twitterin LinkedIn📸 Instagram🔗 Copy link
0:00