Dear Dario, Open Source Claude Code
- 8 minutes read - 1673 wordsPhoto Credit: TechCrunch at https://www.flickr.com/photos/techcrunch/53202070940/
It’s a nightmare that wakes any developer in a cold sweat: I exposed internal code. This week, some poor developer at Anthropic unintentionally published the source code to Claude Code. The internet genies have copied it thousands of times and now all the king’s horses and all the king’s men will never get the code off the internet again.
But now what? I’d argue that Anthropic should open source Claude Code. Here’s the open letter I’d write to Anthropic CEO Dario Amodei if I thought there was any chance he’d pay the likes of me any attention.
Dario,
You’ve had a hell of a week. Pearl tea next time you’re in Manhattan?
It must have been a real rough moment when you learned that the source code to your deeply-loved flagship product, Claude Code, had been released to the internet. It’s a sunny, warm day in Manhattan; I suspect you’re working this weekend in foggy SoMa, San Francisco. I imagine you’re in a war-room now, family neglected, date night canceled. You have my sympathy.
But while your company has been doing the obvious moves (DMCA takedowns, etc.) let’s talk strategy. This moment will look bad in the history books; it is bad. But what you do next can make you a world-historical CEO: a Jobs returns to Apple CEO, a Nadella writes a new era for Microsoft CEO.
Let’s talk assets:
- Developers love Claude Code
- Developers looking at its source are being a little snarky and schadenfreude-ful, but they are looking and that means they think Anthropic worth learning from, worth respecting
- The team that built Claude Code still works for you
You still have great cards to play. While every competitor was shipping VS Code plugins, your team built something developers actually want to live in. And those people — the ones who built that — still work for you.
But here’s my idea: open source Claude Code. Doing so, while leveraging your assets is how you just might be the subject of a visionary leadership brief in the Harvard Business Review.
This will be hard to sell to your investors, but history has some useful analogies.
Mind Share: The Apple Classroom Playbook
Right now you have the eyes of the world on you. That kind of attention is hard to get (I’m sure you’ve reviewed your Marketing budget). But right now you could say:
We think Anthropic will be around a long time. And the secret sauce is this team, our models, and our brilliance at building great experiences that empower developers, like Claude Code. A snapshot of code isn’t fatal to us. In fact, judging by the popularity of our code on GitHub, we think a lot of people want to make Claude Code amazing. So we’re going to make open-source Claude Code a top-tier experience. Like TypeScript or Linux, we’re going to develop this loved product out in the open. We’ll be setting up a new canonical repository and managing our shared endeavor as an exemplar of open source in the twenty-first century.
Instantly hackers from all over the world will start scratching their itch and contributing to a shared endeavor. It’d be the ultimate jujitsu move: you’d take some snarky critical developers and invite them into being collaborators. Personally, I have an unscheduled bugfix request in the Claude Code repo. I’d post a pull request tomorrow if I had a sanctioned code tree to work from.
Interested high-schoolers through PhD candidates will train on your tool (like Minix seeded a generation of hackers before Linus Torvalds built something better).
In short, you could own the AI recruiting pipeline for the next 18 months as your competitors hem and haw. You have no choice, your horse has left the barn. But could you make every bard in the land sing of the beauty of your horse?
And this is the heart of Apple’s iMacs in classroom initiatives and student discounts. Those plucky little machines won the hearts of the doers: the graphics pros, the movie creators, the musicians, the developers (Paul Graham’s “Return of the Mac” memorializes the sea change moment). People may use Windows, but people who create are all pretty much on Macs. I’m sure your internal hardware spend validates this.
Imagine the parallel: Students who learn on — and contribute to — an open Claude Code become the engineers who advocate for it inside their employers or dream of working at Anthropic.
Enterprise Confidence Through Transparency
As you know, and have helped occasion once or twice, enterprises are just not sure what to make of what development in the age of AI looks like. Some of your more cryptic and bowel-loosening utterances — “AI will cause unusually painful disruptions;” “Software engineering automatable within 12 months” — have not been helpful. Some of your company’s stunts (Claude autonomously writing a C compiler; Dario musing that AI would soon replace COBOL migration projects/maintenance entirely) haven’t helped either.
Scaring the bears, pal, was not wise.
But if you open source Claude Code, some of the assets described above make new rain. When an organization can read the source, it is no longer betting on a black box. With your public Claude Code repo you can prove to the enterprise:
AI-enabledClaude Code-enabled developers in huge numbers can ship great features- Claude Code prompted software is as good as gold standard code written by a human
Both claims share one proof: use Claude Code to build Claude Code, out in the open. Every commit becomes evidence; every sprint a case study. That’s dogfooding—and it costs you nothing your team isn’t already doing. But it does need immediate action on your part. The released source has been ribbed heavily by the “never AI contingent.” You need to put effort here, now. And then you need to make this the best repository ever. You need superior—the best—developer advocates to manage the hell out of that Issue queue and sprint management. I suspect you already have that talent on staff, but put your resources here.
You’ll know when your sales engineers / training arm come back and say: “They trust us because they see how we do it, great features hour after hour, 24/7 and they want that.”
Additionally, depending on your license of choice, as Engineering teams and Claude-Code-trained new graduates mingle and adapt Claude like a stem cell into new organs, some great ideas you can’t imagine may ship.
The Linux Precedent
Dario, I feel like I’m seeing history rhyming again, not repeating. I got into Linux in 1993 when I got dozens of 5¼ floppies for a thing called Slackware. By 1999, it was clear to me that Linux was going to be a big thing. At the time, I was working part-time at Lotus (a subsidiary of IBM) after my internship and before graduation. My product (cc:Mail) was being sunset and we were losing customers to Exchange and, to IBM’s preference, Notes.
In that year, as part of my BBA, I argued that IBM should make Linux their premier platform for development. Buy Star Office (now Libre Office) and make it their productivity platform. Bring Notes to Linux with panache. Make Domino server work on Linux. I argued that IBM could hit Microsoft over and over via the CFO’s office: What does Word really get us? Who actually uses all the features of Excel? Businesses no longer paying the MS licensing tax could leverage those savings to compete. IBM did not listen to a 21 year-old me. I did get a good grade. So that was something.
We know what happened, a search engine that didn’t exist as I wrote that turned its propeller-hatted noggin toward that problem, updated some JavaScript patterns and seized that applications niche and made my argument (on the back of their own Linux-chocked data centers). Google Office (or whatever they call it) delivered the features of Microsoft Office you needed for an unbeatable price point.
And about those Linux servers. Linux did grab the cloud pie mightily. In fact, today Windows Azure runs more Linux workloads than Windows ones (go figure, eh?). The company that became the enterprise buddy for Linux was RedHat. Twenty years later IBM would buy RedHat for $34B. In 1999 they IPO’d at $3B. Big Blue missed the vision, missed the buyout, and that was just the start.
In RedHat’s ~20 independent years, IBM spent years watching Linux and RedHat eat into their big iron business. When they finally caught wise, they did so from a position of catch-up, not leadership. The cannibalization had already happened; they were buying back relevance at a premium.
IBM is not even the sharpest example. Microsoft spent a decade fighting Linux. Steve Ballmer called it a “cancer.” And yet today Azure runs more Linux workloads than Windows Server, Microsoft owns GitHub, and they ship their own Linux kernel inside Windows. They did not win by resisting—they won by making themselves indispensable to the ecosystem they once tried to kill.
The pattern is consistent enough to be a law: the companies that try to contain open source movements end up serving them anyway, just later and at greater cost. Anthropic can read the history and skip the painful middle chapter.
Conclusion
So Dario, it’s been a hell of a week. Make some time for loved ones. Take some time for yourself. Yes, this is not what your investors were planning on. But the toothpaste tube has been rolled over by a steamroller — there’s no Escape double-tap to revert state and get it back in the tube. Playing safe is merely meeting ante while playing not to lose. My strategy is a gambit to win. This window will be open but briefly.
This could be a legendary turnaround: Iger at Disney, Gerstner at IBM, Jobs returning to Apple. You could join this trio of turnaround icons. Open source Claude Code with all due haste. Do it before someone else frames the narrative for you.
A Lover of Claude Code,
Steven