The creator of Ruby on Rails just told his engineers to stop writing code—unless the AI breaks.
The Summary
- David Heinemeier Hansson announced at Rails World that 37signals has gone "pencils down" on hand-writing code, with engineers only stepping in when AI agents fail to deliver
- Hand-writing code is now the "exceptional state" at 37signals—the thing you do when the machine breaks, then you "fix the machine"
- A January poll of 150+ software engineers showed 45% feel they're "keeping up" with AI tools, while 17.5% are opting out entirely
The Signal
This isn't a hot take from a VC or an AI company founder. This is DHH—the guy who built Ruby on Rails, who's been writing and thinking about code for two decades. When someone with that track record says "we're done," you pay attention.
The framing matters here. Hansson isn't saying AI will write code someday. He's saying his team has already stopped writing it by hand. Hand-coding is now the exception, the fallback, the thing you do when the AI screws up. Then you don't just fix the output—you "fix the machine." This is a mental model shift: the AI is the primary producer, and human engineers are quality control and trainer.
"This is a recognition of what is already happening."
The gap between what's possible and what people are willing to admit is closing fast. Tools like Cursor, Claude Code, and OpenAI Codex have been in market for over a year. Developers are split: some are staying up late building with agents, others call it lonely and tiresome. The January poll of 150+ engineers shows the distribution: 45% keeping up, 17.5% opting out entirely. The rest are somewhere in between—probably experimenting, uncertain, or quietly falling behind.
What DHH is doing is forcing a reckoning inside his own company. No more "I'll try the AI when I have time" or "I prefer to write it myself." The default is now agent-generated code. If you want to hand-write, you need a reason. That's a management decision, not a technical one. It's culture design.
The real question isn't whether AI can code. It's what happens to engineering teams when the primary skill shifts from writing code to directing agents and debugging their output. DHH's answer: accept it, institutionalize it, and stop pretending the old way is coming back.
The Implication
If you're a software engineer still treating AI coding tools as optional, you're reading the room wrong. The most respected voices in the field are moving past "should we use AI" to "how do we build teams around agents as the default." The companies that standardize on this first will move faster. The engineers who learn to work with agents—not just write code themselves—will stay relevant. The ones opting out are choosing to become more expensive and slower for the same output.
Watch for more companies to formalize this shift in the next six months. "Pencils down" won't stay a 37signals policy. It'll become an industry standard.