What does it mean for software engineering when we no longer write the code? Here's the take from Boris Cherny (Boris Cherny), the creator of Claude Code. Timestamps: 00:00 Intro 11:15 Lessons from Meta 19:46 Joining Anthropic 23:08 The origins of Claude Code 32:55 Boris's Claude Code workflow 36:27... Parallel agents 40:25 Code reviews 47:18 Claude Code's architecture 52:38 Permissions and sandboxing 55:05 Engineering culture at Anthropic 1:05:15 Claude Cowork 1:12:48 Observability and privacy 1:14:45 Agent swarms 1:21:16 LLMs and the printing press analogy 1:30:16 Standout engineer archetypes 1:32:12 What skills still matter for engineers 1:35:24 Book recommendations Brought to you by: • Statsig — ⁠ The unified platform for flags, analytics, experiments, and more. • Sonar – The makers of SonarQube, the industry standard for automated code review. Proactively find and fix issues in real-time with the SonarQube MCP Server: • WorkOS – Everything you need to make your app enterprise ready. Three interesting things from this conversation: 1. Boris automated himself out of code review well before AI. Boris was one of the most prolific code reviewers at Meta company. And he worked hard to minimize time spent on code review. His system::every time he left the same kind of review comment, he logged it in a spreadsheet. Once a pattern hit 3-4 occurrences, he’d write a lint rule to automate it away! 2. PRDs are dead on the Claude Code team: prototypes replaced them. Instead of writing Product Requirement Documents (specs), they build hundreds of working prototypes before shipping a feature. Boris: “There’s just no way we could have shipped this if we started with static mocks and Figma or if we started with a PRD.” 3. This is the year of the generalist (and maybe the year of those with ADHD) Boris’s work has shifted from deep-focus single-threaded coding to managing multiple parallel agents and context-switching rapidly. As Boris put it: “It’s not so much about deep work, it’s about how good I am at context switching and jumping across multiple different contexts very quickly.”show more

Thanks for hosting me on the pod, was a really fun conversation!

Watch or listen on other platforms: • YouTube: • Spotify: • Apple: • Summary and transcript:

@bcherny You don’t just wake up one day and decide to build something like Claude Code. This explains so much.

@bcherny Shipping 100 prototypes for each feature is not something we little people can afford. I guess I'll keep writing specs.

@bcherny It was never 100 prototypes. But I did a walkthrough of the actual ~20 prototypes Boris built for a small feature in two days. It's educational: Less about # of things and more about how much time do you spend on what?

@gustojs @bcherny Every impactful software project I have worked on has been driven by "how much time do you spend on what?". Each one was a straightforward application of the "opportunity cost" I learned about in high school but with more uncertainty, time pressure and cost of failure.

@GergelyOrosz @bcherny That's def a good system to have.

PRDs are dead because building is now cheaper than planning. the old dev loop existed because mistakes were expensive. now you prototype in minutes, ship, learn, iterate. the skill that matters isn't writing code or writing specs — it's taste. knowing what to build and when to stop. the product development loop just collapsed from months to hours.

@bcherny the lint rule approach he used at meta is basically what hooks in claude code do now. automate the pattern, skip the review

The timestamp that matters most: 52:38 (Permissions and sandboxing) and 1:14:45 (Agent swarms). Boris is right that we don't write code. But we haven't solved how we verify what the agents wrote—especially when parallel agents modify shared state. The "generalist" skill isn't context-switching. It's defining deterministic constraints (schemas, types, execution proofs) that agents can't violate, regardless of how many are running in parallel. The review layer shifts from "does this code look right?" to "does this execution have a cryptographic proof of correctness?"

@bcherny the part nobody talks about is that 'not writing code' doesn't mean not thinking about code. if anything you're making architectural decisions 10x faster which means your taste and judgment matter way more than your typing speed ever did

@bcherny I think software engineering is becoming something closer to knowledge curation than coding. Most teams aren't ready for what that means. @ImplicitCloudAI is one attempt at figuring it out.

boris's spreadsheet-to-lint-rule pipeline for code review is the right pattern for agent management too. i track repeated corrections to my openclaw agents the same way — if i give the same steering hint 3+ times, it becomes a permanent instruction in the agent's AGENTS.md. the meta-skill now is noticing your own repetition and automating it away, whether that's lint rules or agent prompts

@bcherny Oh, that's why it is so good in Typescript and it sucks ass in python

@bcherny Very Good! Thank you both, @GergelyOrosz, @bcherny 🙏

Point 3 is the biggest shift nobody's ready for. The "10x engineer" used to mean deep focus on one hard problem. Now it means orchestrating 5 parallel Claude Code agents across different parts of a codebase simultaneously. Boris automated lint rules from repeated code review patterns at Meta — that same instinct (pattern → automate → move up the stack) is exactly what makes someone effective with AI coding agents. The engineers who'll thrive aren't the best typists. They're the best delegators.

@bcherny Been watching AI code generation evolve for years - the shift from "helps you code faster" to "writes the entire feature" is wild. The real question becomes how we validate and maintain what we didn't write ourselves.

@bcherny Claude Code is not just a productivity tool — it is a window into a future! Great interview worth watching. Don't have the time → subscribe and read our structured summaries.

the prototype-over-PRD shift tracks with what I've seen building with Claude Code. a working prototype surfaces edge cases in minutes that a spec would miss entirely. the tricky part is knowing when to stop prototyping and commit — with fast iteration loops it's easy to explore forever and ship never

@bcherny Super cool! We'll need evolutionary thinking more and more as Claude does more and more:

@bcherny the context switching point resonates. used to optimize for deep work blocks but now the skill is holding mental models of 3-4 parallel tasks while agents handle implementation. completely different muscle

@bcherny @grok videoda ne anlatilmak isteniyor.

@bcherny “Now I don’t remember TypeScript at all.” Same.. I ask AI to fix it for me.

the part about Boris building Claude Code as an internal tool first and then realizing engineers outside Anthropic needed it too is the most important lesson here. best dev tools always start as something one person builds to scratch their own itch... not a product roadmap designed by committee. curious how much the system prompt architecture changed once real users started doing things Boris never anticipated

@bcherny @HeyGenLabs summarize

@bcherny Boris dropping truth: prototypes killed PRDs (20 in 2 days for a feature), lint rules automated reviews pre-AI, 2026 is generalist year with ADHD-level context switching. Deep code is dead—managing agents wins. Who's already ditching specs for prototypes?

@bcherny the context switching point is underrated. same shift happening in data collection for frontier models - you're not running one labeling pipeline, you're orchestrating 20 simultaneously. the skill set that wins looks nothing like traditional annotation ops

@bcherny It means we go From doing the work to directing the work

@bcherny @grok 1:35:24 Book recommendations bu bolumu describe et

@bcherny Same question is hitting design and creative production right now. When AI generates the first draft, the job shifts from making to editing and judging. The taste does not go away, it just moves earlier in the process.

@bcherny automating code review while building systems that need miracles to debug. senior devs learned intuition breaking things at 2am. next gen has taste, no clue why lights go out.

@bcherny the context switching point is underrated. the best devs I've seen using Claude Code aren't the ones who write the most detailed prompts. they're the ones who can hold several parallel threads at once, hand off, review, redirect. it's a different muscle than writing code.

@bcherny PRDs are dead, prototypes replaced them. This hit home. I've been writing specs as plain English briefs and feeding them straight to Claude Code. Faster than any PRD process I ran as a PM

@bcherny 80% AI-generated code still needs someone to decide which 80% to write and which 20% to kill. Architecture judgment doesn't autocomplete. Genuinely curious — what does code review look like when most PRs aren't written by humans?

@bcherny The age of the context switcher. Who would have thought.

@bcherny And I still have the book lolz Bought it in 2025 to master typescript so as to be in a position to build better softwares And yes it’s worth it even with AI. What I learn transformed how I want AI to work for me

@bcherny Impressive.