Steam rising from a San Francisco coffee shop laptop, Boris Cherny hits ‘post’ on X, unleashing a thread that has 10,000 developers scrambling to mimic his terminal.
That’s the scene last week. Cherny, head of Claude Code at Anthropic, didn’t just share config files—he handed out a blueprint for coding at warp speed. And yeah, the hype’s real this time.
Look, in a market where OpenAI’s burning billions on data centers and Nvidia’s stock rides the AI wave, Anthropic’s playing chess, not checkers. Cherny’s setup proves it: one engineer, leveraging Claude’s smarts, outputs like a 20-person team. Market dynamics? Anthropic’s valuation just ticked up on rumors of enterprise deals— this workflow’s the secret sauce.
Developers aren’t just buzzing; they’re converting. “If you’re not reading the Claude Code best practices straight from its creator, you’re behind as a programmer,” Jeff Tang tweeted, capturing the frenzy.
Why Does Boris Cherny Run Five Claudes in Parallel?
Short answer: because linear coding’s dead.
Cherny doesn’t peck at keys in a single VS Code window. No—he’s got five Claude instances firing in iTerm2 tabs, numbered 1-5, with macOS notifications pinging when one needs a nudge. One agent’s grinding tests. Another’s refactoring that crusty legacy code. A third? Drafting docs while you sip coffee.
And get this—he juggles 5-10 more on claude.ai in the browser, teleporting sessions local with a custom command. It’s not multitasking; it’s a real-time strategy game. Remember Starcraft? Users are already saying it feels exactly like that—command units, not micromanage zerglings.
Here’s Cherny himself:
“I run 5 Claudes in parallel in my terminal. I number my tabs 1-5, and use system notifications to know when a Claude needs input.”
This isn’t gimmickry. Anthropic’s president, Daniela Amodei, just preached ‘do more with less’ in a CNBC spot—Cherny’s living it. While OpenAI scales hardware to the moon, Anthropic orchestrates smarter. Result? Exponential gains without the capex hangover.
My take: this echoes the 1970s Unix revolution. Back then, small, composable tools beat bloated mainframes. Today, Cherny’s fleet of Claudes does the same to IDEs. Bold prediction—within a year, dev productivity metrics at Anthropic-using firms will lap competitors by 3x.
Is the Slowest Claude Model Secretly the Fastest?
Counterintuitive as hell.
Everyone chases latency—faster tokens, snappier chats. Cherny? He picks Opus 4.5, Anthropic’s beefiest, slowest beast. With ‘thinking’ mode cranked on for everything.
Why? Smarter AI means fewer fixes. “It’s the best coding model I’ve ever used, and even though it’s bigger & slower than Sonnet, since you have to steer it less and it’s better at tool use, it is almost always faster than using a smaller model in the end,” he posted.
CLAUDE.md: The File That Makes AI Learn from Screw-Ups
AI’s got amnesia—bad. Cherny’s fix? One git-tracked file: CLAUDE.md.
Every Claude flub—wrong style, dumb architecture call—gets logged there. Next session, boom, Claude reads it, doesn’t repeat. It’s turning the repo into a self-healing beast, no fine-tuning needed.
This scales team-wide. No more ‘wait, why does Claude hate our linter?’ One file, permanent lessons. Genius in simplicity.
But here’s my sharp edge: Anthropic’s PR spins this as magic, yet it’s engineering basics amplified by better models. OpenAI’s chasing agent swarms with o1-preview hype—cute, but Cherny’s already field-testing the real deal. If they don’t copy this fast, they’ll eat dust in dev tools.
Does This Crush Traditional Dev Teams?
Market ripple: yes.
Enterprise CTOs, listen up. Headcount’s your biggest line item—Cherny’s workflow slashes it. One dev + Claude fleet = department output. Gartner’ll soon chart this; early adopters like Midjourney (Anthropic users) are already humming.
Skeptical? I ran the numbers. Assume Opus 4.5 at 10x Sonnet’s compute cost, but 5x fewer iterations. Net: 2x faster end-to-end. Stack five agents? You’re at 10x a solo human. Small teams become juggernauts.
Critique time—Anthropic’s not immune to hype. ‘World’s most advanced coding agent’? Prove it in benchmarks, not tweets. Still, data says they’re onto something: Claude’s tool-use laps GPT-4o in coding suites.
And the parallel no one’s drawing? It’s like the spreadsheet killing ledgers in the ’80s. VisiCalc didn’t add brains; it orchestrated human intent better. Claude Code does that for code.
Why Anthropic Might Own Agentic Coding
Forget trillion-param races. Orchestration wins.
OpenAI’s infra binge—$100B+ projected—funds raw scale. Anthropic? Lean, constitutional AI focus yields stickier tools. Cherny’s thread validates it: superior models + human-AI symbiosis > brute force.
Prediction: By Q4 2025, Claude Code captures 40% enterprise dev market. OpenAI pivots to catch up.
Developers losing minds? Smart money’s copying now.
🧬 Related Insights
- Read more: OpenClaw’s Rocket Ride: From GitHub Obscurity to Nvidia’s Agentic AI Obsession
- Read more: SK hynix’s AI Memory Dominance: Record Profits and a Full-Stack Bet That Could Reshape DRAM Wars
Frequently Asked Questions
What is Boris Cherny’s Claude Code workflow?
He runs five parallel Claude instances in iTerm2 tabs with notifications, plus browser sessions, using Opus 4.5 exclusively and a CLAUDE.md file for fixes.
Does Claude Code workflow boost developer productivity?
Absolutely—users report small-team output from one person, like commanding AI agents in a strategy game.
Is Anthropic’s Opus 4.5 better for coding than faster models?
Yes, per Cherny: less steering needed means overall faster, despite higher latency.
How do I set up CLAUDE.md for my team?
Add it to your git repo, log every AI mistake there—Claude reads it next time for consistent style.
