DHH's new way of writing code

DHH's transformation wasn't philosophical—it was empirical. On Lex Fridman's podcast six months earlier, he had dismissed AI autocomplete tools as infuriating, like someone constantly interrupting mid-sentence. The models weren't good enough, and the ergonomics were terrible. Then Anthropic released Claude Opus 4.5 on November 27, 2024, and everything shifted.

The combination of a frontier model that could consistently produce merge-worthy code and agent harnesses that gave AI access to terminal tools created a new paradigm. DHH describes it less like managing a team and more like stepping into a «super mech suit» where he suddenly has twelve arms and can look at seven screens simultaneously. He's still the one doing the work—just hyper-accelerated.

Within weeks, DHH went from code-first (opening his editor, writing code, occasionally consulting AI) to agent-first (starting every project by instructing agents, reviewing their output, and making alterations only when needed). The speed of this shift shocked even him. What took months or years to contemplate—like optimizing the fastest 1% of requests—now happens in days as side projects.

DHH has always believed that aesthetics and correctness are inseparable. «When something is beautiful, it's likely to be correct,» he says, citing mathematics, physics, and software as domains where this principle holds. This philosophy is why he fell in love with Ruby—it produces the most beautiful code in his view, with an elegant balance of expressiveness and pragmatism.

In the age of AI agents, this philosophy becomes even more critical. Agents can generate working code, but can they generate beautiful, maintainable, coherent code that fits the grain of the system? Not yet—at least not without supervision. DHH won't merge sloppy agent output any more than he'd merge sloppy code from a junior developer. His bar remains sky-high, and that bar is what makes agent acceleration work for him.

This is why DHH believes standout designers and engineers who care deeply about craft will become more in demand, not less. As implementation gets commoditized, taste, judgment, and the ability to shape systems with intention become the differentiators. The constraint is no longer «can we build it?» but «should we build it, and what's the right way?» That's a deeply human question, and one that requires aesthetic sensibility as much as technical skill.

Most companies treat designers as people who make specs look pretty. At 37signals, designers are product managers, implementers, and decision-makers rolled into one. They figure out what should be built, how it should work, and increasingly—with agent acceleration—how to implement it in CSS, HTML, and even JavaScript or Ruby.

DHH argues that when you combine these three roles—product thinking, visual design, and implementation—you get someone who understands the materials they're working with. They know how CSS stretches, how HTML structures, how the web wants to behave. It's like a jewelry designer knowing the properties of gold or an architect understanding load-bearing structures. Working natively in the medium produces better, more coherent results.

With agents, this model is becoming even more powerful. Designers at 37signals can now use AI to produce full working prototypes, not just as they'll be merged, but as a demonstration of the final shape and interaction. DHH recently hired a designer, Zoltan, who embodies this philosophy—someone who can think, design, and build fluidly across the entire stack. DHH believes this convergence of skills, now accelerated by AI, is the future of how great software gets made.

DHH has been financially independent for years, yet he works as hard as ever. Why? Because he genuinely loves computers. He's been obsessed since age five, whether playing video games, building software, or tinkering with Linux distributions. For him, wealth was never a checkpoint or a finish line—it was a byproduct of doing work he found deeply satisfying.

He rejects the notion that leisure is the ultimate goal. Psychological research consistently shows that purpose and mission are what make people happy, not passive consumption of free time. DHH has seen this pattern play out with every entrepreneur who sells their company, sits on a beach for three weeks, and then immediately jumps back into building something new.

Right now, DHH is more excited about computers than he's been in decades. Agent acceleration feels like the early 2000s when he first discovered Ruby—a time of pure creative possibility. He wakes up every morning intensely curious about what's happening in AI, what's possible today that wasn't yesterday. And while he's careful not to sacrifice sleep, health, or family time, he's leaning in hard. The mission remains the same: make things with computers, make them beautiful, and share them with the world.