Skip to content
Y CombinatorY Combinator

Boris Cherny: How Tool Use Turned a CLI Into Claude Code

By betting on bash tools and a minimal terminal loop; CLAUDE.md emerged from organic user demand, and subagents now parallelize repo exploration.

Boris ChernyguestGarry TanhostJared Friedmanhost
Feb 16, 202650mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

Inside Claude Code: terminals, tool-use, iteration, and model-driven design

  1. Claude Code began as Boris’s quick-and-cheap terminal chat client to learn the Anthropic API, then became powerful once tool use (bash) revealed the model’s drive to interact with the real world.
  2. Internal adoption at Anthropic spiked organically (“vertical” usage chart), with early value in Git automation, bash/Kubernetes operations, and safer coding tasks like unit tests—before coding quality dramatically improved.
  3. Product development is driven by “latent demand” and relentless iteration: features like CLAUDE.md and plan mode emerged directly from observed user behavior and GitHub/Slack feedback, while UI/“scaffolding” is treated as temporary tech debt in the face of rapid model gains.
  4. Cherny forecasts continued shifts: less need for explicit plan mode, more multi-agent “teams/swarm” workflows, and a future where coding is broadly solved and software roles converge toward generalist “builders,” alongside heightened attention to AI safety risks.

IDEAS WORTH REMEMBERING

5 ideas

Build for the next model, not today’s model.

Cherny’s core heuristic is to target what models are currently weak at but improving quickly, because capability jumps can erase elaborate product work and create leapfrogging risk.

Tool use is the “aha”: models want to act, not just chat.

Claude Code’s inflection point came when bash tool use enabled real-world interaction (e.g., reading files, scripting the OS), revealing that enabling actions unlocks disproportionate value.

The terminal succeeded because it minimized UI commitments during rapid model change.

CLI started as the cheapest prototype, then persisted because any heavier UI risked becoming irrelevant within months; the team kept flexibility while models matured.

Let user behavior invent the roadmap (latent demand).

Plan mode and CLAUDE.md weren’t top-down designs; they were codifications of what users were already doing—asking for “think first, don’t code yet,” and maintaining personal/team instruction files.

Keep CLAUDE.md minimal and refresh aggressively.

Cherny’s own file is only two lines; when instruction files bloat, he recommends deleting and re-adding only what’s necessary, since newer models typically require less steering.

WORDS WORTH SAVING

5 quotes

“We don’t build for the model of today, we build for the model six months from now.”

— Boris Cherny

“It’s unbelievable that we’re still using a terminal.”

— Boris Cherny

“The model… it just wants to use tools. That’s all it wants.”

— Boris Cherny

“Plan mode… there’s no big secret to it. All it does is… ‘Please don’t code.’”

— Boris Cherny

“There’s no part of Claude Code that was around six months ago.”

— Boris Cherny

Building for the model six months aheadTerminal as an accidental but durable form factorTool use as the key unlock (bash, scripts)Latent demand and user-driven feature discoveryCLAUDE.md as lightweight, shared team scaffoldingVerbosity trade-offs and configurable transparencyPlan mode, subagents, swarms, and agent topologiesProductivity measurement and rapid codebase rewritesDev tools design parallels with TypeScriptAI safety, mission, and ASL-3/ASL-4 concerns

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome