Boris Cherny

Sources#

• A Field Guide to Fable: Finding Your Unknowns
• Anthropic's Boris Cherny: Why Coding Is Solved, and What Comes Next
• Boris Cherny: We Cut 80% of Claude Code's Prompt
• How Anthropic's product team moves faster than anyone else | Cat Wu (Head of Product, Claude Code)
• Loop Engineering
• Thread by @AndrewYNg

Summary#

Creator and tech lead of Claude Code at Anthropic. Engineer-by-background, author of Programming TypeScript. Joined Anthropic late 2024 via internal incubator "Anthropic Labs" (the same incubator that produced Claude Code, MCP, and the Anthropic desktop app, then disbanded — re-formed under Mike Krieger for "round two" by mid-2026). Most-listened guest in Lenny's Podcast history at time of Cat Wu's appearance (April 2026).

Notable claims and stances#

• Coding is solved (for him). Writes 100% of his code via Claude Code; has logged days of 150 PRs. Caveat: not yet true for very large codebases or off-distribution languages.
• Phone-driven workflow. Most work now done from the iOS Claude app; runs ~5–10 sessions concurrently, "a few hundred" agents during day, "a few thousand" overnight.
• Loops as the future. Heavy user of /loop (Claude-scheduled cron jobs) for PR babysitting, CI healing, Twitter feedback clustering. Sees the Agent Loop Pattern and parallel-agent fan-out as the next harness primitive (see Harness Shrinkage as Models Improve). By June 2026 the stance had sharpened into the thesis of Loop Engineering: "I don't prompt Claude anymore. I have loops running that prompt Claude and figuring out what to do. My job is to write loops."
• Build for the next model. Built Claude Code in late 2024 knowing it had no PMF; PMF arrived with Opus 4 in May 2025 and inflected with each subsequent release (4.5 → 4.6 → 4.7).
• Generalists across disciplines. Every member of the Claude Code team writes code — including the engineering manager, PM, designers, data scientist, finance, user researcher (see Engineer PM Convergence).
• Printing press analogy. Sees Printing Press Software Democratization as the right historical parallel for AI-assisted software building.
• The model as organism, not system. (YC interview, July 2026) Building on models is "so different than all the engineering I've ever done" — no upfront system design; the model is "almost like a living creature… every model generation it behaves differently, it has a slightly different personality," so the discipline is empirical: try, observe, iterate. Rejects one-weird-trick culture ("don't listen to the LinkedIn influencers") and names over-specification as the veteran engineer's characteristic failure mode — describe the task, guardrails, and exit criteria, "then let the model cook" (Vibe Coding vs. Agentic Engineering).
• Verification as the elicitation key. "Give it a task that's too hard, give it the tools to verify the work like you would yourself" — verification is "probably the single most important thing that people do not get right." His running demo: a Claude-in-Slack session rewriting the Electron desktop app in Swift, verifying by screenshotting both apps in a macOS VM and comparing pixel by pixel — still running after 2+ weeks, live-blogging its own progress screenshots to a Slack channel it created.
• Coding-is-solved caveats (updated). Still excludes deep systems code, distributed systems, and pixel-level UI verification; his audience polls (YC, July 2026) found a large minority at 100% agent-written code, ~half at >50%.

Background#

First learned to program on a TI-83 calculator in middle school — a BASIC algebra solver to cheat on math tests, later rewritten in assembly when calculus outgrew BASIC; wrote an early internet guide to TI-83 programming. His stated advice for CS students follows from it: learn by building practical things — design sense, business sense, talking to users — not theory alone.

Working style (from Cat Wu's account)#

Cat Wu (Head of Product for Claude Code and Cowork) describes Boris as "the product visionary" setting a 3–6 month vision; she translates it into a near-term path and handles cross-functional alignment. They estimate 80% mind-meld with the remaining 20% split between domains they each drive. The split is "remarkably blurry" — typical of Engineer PM Convergence inside the team.

Key quotes#

• "Loops are the future at this point. If you haven't experimented with it, highly highly recommend it."
• "Software will be a thing that is fully democratized, that anyone can do… The best person to write accounting software, I think maybe even today, is not an engineer, it's a really good accountant."
• "If you're starting fresh, then you can kind of build with AI natively from the ground up."

Connections#

• Claude Code — created and leads the product
• Cat Wu — Head of Product partner
• Anthropic — employer
• Agent Loop Pattern — heaviest user, primary advocate
• Loop Engineering — "my job is to write loops" is one of the two practitioner quotes the discipline is named on
• Engineer PM Convergence — describes own team as the prototype
• Seven Powers Applied to AI — Acquired-listener take on which moats survive AI
• Printing Press Software Democratization — his framing
• Harness Shrinkage as Models Improve — predicts Claude Code "may be 100 lines of code a year from now"; the YC interview supplies the methodology (ablation, SIMPLE=1, more-intelligent-without-prompts)
• Dynamic Workflows: An Algebra for Agents — designed the "algebra for agents" (his functional-programming background made product); his stage account of the Bun rewrite is now partly superseded by Jarred Sumner's first-party methodology — the production claim held, "one prompt, one dynamic workflow" and ">100k LOC" did not
• Latent Capability Overhang — coined the product-side framing: "hobbling" and "product overhang"
• Compute Allocator — "coding is solved (for me)" is the precondition; Thariq Shihipar's allocator framing names the role that remains once it's true
• Building Is Cheap, Arguing Is Expensive — debated refactoring approaches with Fiona Fung by generating three real PRs instead of whiteboarding
• Vibe Coding vs. Agentic Engineering — Karpathy's parallel "coding is solved" arc; both treat themselves as the case study
• The Three Loops of AI-Native Building — Andrew Ng credits him (with Peter Steinberger) for making "loop engineering" a buzzphrase, then bounds it: his loops are the innermost of three
• Unknowns as the Agentic Bottleneck — Thariq Shihipar names him as the low-unknowns exemplar: "it is obvious to me that they know what they want in-detail… but they also assume unknowns"

Sources#

• Anthropic's Boris Cherny: Why Coding Is Solved, and What Comes Next — Sequoia AI Ascent 2026 interview
• Boris Cherny: We Cut 80% of Claude Code's Prompt — Y Combinator interview (2026-07-27, practitioner-opinion): ablation methodology, model-as-organism stance, verification-as-elicitation, the Electron→Swift Slack experiment, TI-83 backstory
• How Anthropic's product team moves faster than anyone else | Cat Wu (Head of Product, Claude Code) — Lenny's Podcast (referenced as collaborator)
• Thread by @AndrewYNg — credited (with Steinberger) for popularizing "loop engineering" (practitioner-opinion)
• A Field Guide to Fable: Finding Your Unknowns — cited as an exemplar of the agentic coder with few unknowns (practitioner-opinion)