Source-code moats are dead; believers in them are deluding themselves
DHH argues agents drive software development costs to zero. Mac and Windows' mouse-driven interfaces are fundamentally incompatible with agent workflows. Linux's historically criticized features—terminal commands, complex configs, verbose error messages—are exactly what agents need.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Graphical operating-system settings are fundamentally incompatible with agent workflows
Remapping Mac workspace shortcuts required manually clicking through settings menus—agents can't batch-configure the way GUIs force. DHH demonstrated letting an agent modify his running macOS directly, and it worked. After five minutes of that experience, he says you can't go back. Mac and Windows' forty years of mouse-click interaction doesn't fit the agent era.
— David Heinemeier HanssonLinux's historical flaws are exactly what agents want most
Terminal dependency, complex config files, obscure error messages—Linux's traditional criticisms. DHH sees them as exactly what agents need: command-line interfaces, predictable editable configs, detailed error information. This year's inversion: the historical "flaws" have become reasons to choose Linux if you're working with AI.
— David Heinemeier HanssonEcosystem lock-in through scarcity breaks when anyone can code
An Omachi community member replaced a Mac-only Lightroom dependency by implementing a better version in two weeks—after Adobe spent over a decade and tens of millions developing it. This shows ecosystem lock-in's keystone is crumbling: application availability once locked you in, but now anyone with agent access can generate software in minutes. This resets mobile competition dynamics entirely.
— David Heinemeier HanssonToken throughput speed determines whether software feels alive
Omachi currently runs at 50–120 tokens per second—below the threshold for smooth interaction. Fireworks recently demonstrated 1500 tokens per second, generating a playable Tetris in four seconds and reshaping it in real time as DHH shouted modifications. As token throughput scales to hundreds of thousands and millions per second in the coming years, software will become continuously self-rewriting in real time.
— David Heinemeier HanssonClaiming source code as a moat is now self-delusion
DHH compares Omachi's development progress over months with the gap between Opus 4.8 and 5.5, noting this pace is accelerating and new models are likely by Christmas or early next year. His conclusion: anyone who thinks source code—even millions of lines—protects their ecosystem is deluding themselves.
— David Heinemeier HanssonEven unlimited token budgets need only three subscriptions in practice
After twenty years of building wealth, DHH can afford unlimited token spending. Over three months, only two weeks came close to maxing his quota. Most work runs on a single $200 20X subscription to Claude—frontier models, not open-source alternatives. And he maintains just three subscriptions total. His point: access isn't confined to those with infinite budgets.
— David Heinemeier HanssonBeing in the game matters even when trailing by six months
DHH acknowledges that open-source models like GLM and DeepSeek likely lag by six or seven months; many benchmarks are suspect anyway, and they may not keep up with Opus in real-world scenarios. But their existence matters: they run locally on your hardware, they preserve competition, and they prevent two labs from owning the market and inflating token prices. Without them, we'd repeat Google and Meta's capture of 99% of internet advertising value.
— David Heinemeier HanssonThree years to surpass macOS's global installed base is realistic
Omachi began last year as DHH's hand-written bash-script hobby—just better computers for him and a few friends. With agents, his ambition exploded: beat macOS globally within three years—180 million total users, which doesn't even rank as a tier-C internet service. He even questions whether twelve months is feasible; his reasoning: if labs truly believe recursive self-improving AI arrives within twelve months, an OS's market share should be surmountable.
— David Heinemeier HanssonIn their own words · checked verbatim
Every single thing that made Linux hard to use in the past, oh, it's very terminal heavy, it's got a bunch of config files that's hard to figure out, and it's got inscrutable error messages. These are the three things agents love.
David Heinemeier Hansson7:20
anyone who still thinks that there is moat in source code, even millions of lines of it, or that that is going to protect their ecosystem, are delusional.
David Heinemeier Hansson15:34
I am blessed with the entire dopamine payoff in the last 2%. I don't actually give a fuck about the first 98%.
David Heinemeier Hansson19:38
Do you know how many tokens I buy? I got three clawed subscriptions. That's it.
David Heinemeier Hansson20:40
GLM53 or DeepSeek Flash, I don't know, six, seven months behind-ish on some tasks, not even.
David Heinemeier Hansson28:53
There's only 180 million Mac OS users in the world. The Mac is not that big.
David Heinemeier Hansson34:59
The future is already here. It's just not evenly distributed.
David Heinemeier Hansson37:59
I love this energy that now individuals, thanks to the power of AI, can actually play a pivotal role in the forward progression of the industry.
David Heinemeier Hansson46:16
Figures
| Fireworks demo token throughput | 1500 tokens per second | 13:33 |
| macOS global users | 180 million | 34:59 |
| Omicron Foundation funding | ~$23 million | 11:29 |
| DHH's Claude subscriptions | $200 20X tier, 3 total | 20:40 |
Glossary
- Omachi
- AI-first Linux distribution led by DHH, with core code generated by agents.
- Omicron Foundation
- Nonprofit organization operating the Omachi project.
- vibe code
- Writing software by describing intent to AI rather than hand-coding.
- pseudo-anonymous hacker
- Developer active in open communities using a pseudonym instead of their real identity.
How to listen
Founders wondering whether agents will reopen competition in operating systems and desktop markets; engineers and investors writing code with AI and questioning whether a source-code moat still holds.
Skip the opening self-introductions and the closing tangents about offices, nineties demo culture, and dial-up tones—information-light.