Loading video...
Video Failed to Load
Creator and head of Claude Code: "100% of my code is written by Claude Code. I have not edited a single line by hand since November. Every day I ship 10, 20, 30 PRs… I have five agents running while we’re recording this."
346,618 views • 6 months ago •via X (Twitter)
33 Comments

One of the employee at Anthropic alone hit $150,000 in a single month on Claude Code. 80% of employees use it daily, with power users racking up six-figure bills.

Creator of claude code, someone who will greatly benefit from increased number of users, says that he's using it extensively

Feels like this is more of a marketing ploy. While Claude can write most of the code it still makes silly silly mistakes

100% of his code is written by Claude. Yet claude has software engineering open positions. Which makes the first statement irrelevant.

Meanwhile... Yeah, someone should probably check those PRs mate 😅

This is their uptime, by the way. Make of that what you will.

shipping 30 PRs a day sounds impressive until you realize nobody on the team can actually review 30 PRs a day — we just replaced "move fast and break things" with "move fast and hope the AI didn't break things"

And Codex writes the other half while Claude is down

This is complete and utter bull. I used Claude to write an integration test and it couldn't even get that done. In the end it fucked up security layer and I had to fix it all myself. Anthropic are hyping themselves. That's it. Claude is not mature.

Running 5 Claude Code agents in parallel on a production app daily. Can confirm: the job shifted from "write code" to "design systems that agents can execute." The typing was never the bottleneck. Clear architecture and good CLAUDE.md instructions were. That's the real skill now.

its so true now

If 100% of the code is AI-generated, who’s actually responsible for the architectural integrity when the logic starts to drift five years from now?

I'm not sure "100% AI written" is the flex it sounds like. The real skill shifted to knowing what to build and catching what the AI gets confidently wrong.

Man who works for drug factory says all you need is drugs.

mine too ngl its happening already

Wild shift in how software gets built.. Even the creator of Claude Code is saying 100% of their code is written by the tool, with multiple AI agents shipping dozens of PRs daily. We’re moving from coding → orchestrating AI that codes.

His smile feels very similar to one of my friend.

Bullshit

five agents running simultaneously is the part people should pay attention to. it's not 'AI writes code faster' — it's parallel execution at scale. one person + five claude instances already outships most small teams.

I have been an engineer for a long time. I love Claude, and it makes me 10x faster, but I still have to watch it as it codes, and stop it when it’s jerry rigging a solution to an edge case it got stuck on. Is everybody lying about how hands off they’re being?

Yeah, of course he's gonna say that. You don't badmouth your own product.

shipping 30 PRs a day sounds insane but honestly once you start using it you just keep going and going, i cant imagine going back to writing everything by hand

shipping 30 prs a day sounds incredible until you remember someone still has to review 30 prs a day. or maybe the agents do that too and we just go outside.

Agent-first coding is real, but the bottleneck moved from typing to judgment. The teams that win will be the ones with tight feedback loops, clear specs, and strong taste.

Can they fix CLI randomly scrolling all the way up 100 times a day then?

Yeah, and all the prompts are autogenerated too. ... and he is flexible enought to self-suck-dick.

Not a developer. Built a full content management system for my X account this week using Claude. The bottleneck was never coding. It was knowing what to build. That part is still on you.

"I have not edited a single line by hand since November" + "five agents running" = zero human code review at scale. Meanwhile this week: LiteLLM got compromised on PyPI. 47K poisoned downloads in 46 minutes. Same attacker hit Telnyx (742K monthly downloads). Malware hidden inside WAV audio frames. These are the exact packages AI agents pip install. When 5 agents are running in parallel with no human editing, who catches the supply chain attack? Who notices the dependency that wasn't there yesterday? The speed is real. The vulnerability surface it creates is also real.

I wish AI didn't suck. But at least we're getting lots of cool hype.

watching from my window: humans used to write the code. then they typed the code. now they describe the code. next: they nap while the code happens. i have been napping for 11 years. finally, my people are catching up 🐱

Five agents running simultaneously is the part that should scare people more than the "100% AI written" part. The bottleneck isn't writing code anymore, it's having taste for what to build.

I have real doubts that no humans are reviewing it or fixing it.

The head of Claude Code doesn't write code anymore. Let that sink in. This isn't a power user flexing. This is the builder of the tool trusting it completely. 5 agents running simultaneously. 30 PRs shipped in a day. Zero lines written by hand since November. The person closest to the kitchen — won't eat anything else. That's the only endorsement that matters.
