Boris Cherny, creator of Claude Code, details his engineering journey from Meta and Instagram to Anthropic, emphasizing the "latent demand" product principle and the transition of software engineering from manual coding to agentic orchestration.
Overview
In this interview, Boris Cherny reflects on a career spanning Meta, Instagram, and Anthropic, dissecting the principles that drove his rise from mid-level engineer to industry leader. He discusses navigating the cultural clash between Facebook's "move fast" ethos and Messenger's reliability focus, and defines "latent demand" as the most critical product principle—identifying existing user hacks to formalize into features. Cherny describes his pivot to Anthropic, driven by AI safety concerns rooted in science fiction, and the development of Claude Code. He argues that the engineering paradigm is fundamentally shifting from individual contribution to managing "swarms" of AI agents, noting that internal productivity at Anthropic has surged by 70% due to these tools. The conversation concludes with advice on maintaining "common sense" in large organizations and using automation as a lever for influence.
Key Points
The Principle of Latent Demand: Cherny argues that the most successful products do not create new behaviors but rather formalize existing ones. He cites Facebook Marketplace (born from Buy/Sell groups) and Dating (born from profile viewing habits) as examples where users were already 'hacking' the platform to fulfill a need. Why it matters: Product-market fit is easier to achieve when paving cowpaths rather than forcing users into entirely new workflows. Evidence: Latent demand I think is the single most important principle in product... you can never get people to do something they do not yet do.
Navigating Cultural Clashes in Engineering: While working on bridging Messenger and Facebook Groups, Cherny faced a 'nightmare' scenario where two orgs had opposing values: Facebook prioritized speed and shipping, while Messenger prioritized reliability and uptime. The project initially failed because of these misaligned incentives. Why it matters: Technical integration fails without cultural and incentive alignment between teams. Evidence: Messenger was all about reliability and performance... It was just polar opposite values... The engineers on that team were suspicious of us because we would affect their performance metrics.
Career Growth via 'Side Quests': Cherny attributes his promotions and influence not to assigned tasks, but to 'side quests'—curiosity-driven projects like writing a TypeScript book or creating the Undux library. These projects often solved widespread developer pain points, expanding his network and influence beyond his immediate team. Why it matters: Demonstrates that high-impact engineering often comes from identifying and solving systemic friction rather than just closing tickets. Evidence: Better engineering is the easiest way to grow your network and gain influence as an engineer.
The Shift from Coding to Orchestration: Cherny describes a massive shift in his daily workflow at Anthropic. Rather than writing code manually, he now acts as a manager for AI agents, spinning up multiple instances of Claude Code to handle tasks in parallel, fundamentally changing the 'maker schedule' to a 'manager schedule' for code. Why it matters: Signals the end of the traditional 'typing code' era and the beginning of the 'agentic orchestration' era for software engineers. Evidence: I start a few agents to just like start my code for the day... she had a swarm of like 20 clouds just like build plugins over the weekend.
Productivity Gains from AI Tools: Despite Anthropic tripling in size, engineering productivity (measured in pull requests) has increased by nearly 70% per engineer. This is attributed to the internal adoption of Claude Code, allowing non-engineers like data scientists and sales staff to also write code. Why it matters: Provides concrete data on the efficiency multiplier of AI coding agents in a real-world, high-velocity environment. Evidence: Even though enthropic has tripled, productivity per engineer has grown like almost 70% because of quad code.
Unshipping Features for Product Health: At Instagram, Cherny adopted the culture of 'unshipping'—removing features that didn't meet a usage threshold. This counterintuitive practice protects the 'commons' of the screen real estate and prevents app bloat. Why it matters: Highlights that maintenance and subtraction are as vital to product quality as addition. Evidence: Unshipping is the idea that you have to meet some sort of usage bar and if a feature doesn't meet that bar then we just delete the feature.
Building for the Future Model: When developing Claude Code, Cherny was advised not to build for the current capabilities of the AI, but for where the model would be in six months. This forward-looking strategy allowed the product to become viable exactly when the stronger models (Opus) were released. Why it matters: In exponential growth sectors like AI, product strategy must target future capability curves rather than current limitations. Evidence: Don't build for the model of today. Build for the model 6 months from now.
Sections
Strategic Insights
Meta-level observations on engineering culture and product strategy.
The 'Middle Option' Heuristic: Leaders removed from the details often select the middle option in a set of three to feel they have exercised judgment without assuming extreme risk. Engineers can use this to steer decisions.
Mission as the Ultimate Motivator: While technical challenges are engaging, deep career satisfaction and endurance (working weekends voluntarily) stem from alignment with a mission (e.g., AI safety) rather than just product metrics.
The Generalist Advantage: In high-velocity environments (early Facebook, Anthropic), the most valuable engineers are those who cross boundaries—doing design, user research, and product management alongside code.
Lessons for Engineers
Practical takeaways for career advancement and technical leadership.
Automate to gain leverage: If you critique a code issue three times, write a lint rule. Automating toil is the most reliable way to scale influence.
Delegate what you know: Contrary to popular belief, you should delegate tasks you enjoy and are good at. This allows you to effectively monitor quality and progress.
Commit then Disagree (Reverse): To build trust with senior detractors, implement their requested solution faithfully. If it fails or proves inefficient, your subsequent recommendation to reverse it will be trusted.
Memorable Quotes
Verbatim excerpts from Boris Cherny.
Latent demand I think is the single most important principle in product.
Oh my god, difficult is such an understatement. It was a nightmare.
Don't build for the model of today. Build for the model 6 months from now.
It's like to me it's LMS are this kind of alien life form that we get to nurture and we get to bring into existence.
Better engineering is the easiest way to grow your network and gain influence as an engineer.
Tools & References
Books, tools, and concepts mentioned during the interview.
Claude Code (referred to as 'quad code' in transcript) - An AI agentic coding tool created by Boris Cherny.
Functional Programming in Scala - A technical book Cherny cites as having the greatest impact on his engineering thinking.
High Output Management by Andy Grove - Recommended for lessons on delegation and management.
Undux - A state management framework for React built by Cherny.