David Heinemeier Hansson opened Rails World 2026 not with a vision for Rails, but with an announcement that he's moved on. He declared himself retired from professional programming, now a "maker" who lets LLMs generate code he doesn't read. His flagship product Hey is being rewritten in Rust with native apps on every platform. Only 3% of his output this year was Ruby. The opening keynote of the world's premier Rails conference was, functionally, a farewell. The numbers DHH presented don't survive contact with his own caveats. He claims 150,000 lines of code in August versus a historical average of 30,000 lines per year, then immediately concedes the comparison is unfair — the Rust output is verbose, LLM-generated slop he wouldn't tolerate in hand-written Ruby. The Hey Next performance gains (99% less CPU, 95% less memory) are attributed to Rust and agents, but stripping the web frontend from any stack would yield massive savings. Which gains come from Rust and which from dropping the web app is, as critic Jared Norman notes, unknowable. The intellectual framework is riddled with contradictions. DHH argues humans should never read code, then acknowledges security is a looming concern — fifteen minutes apart in the same talk. He insists UI matters enough to justify rewriting Hey as six native apps, then demands every service expose a CLI for agents. He cites a productivity study to claim 1000x developer multipliers, but the original paper measured tooling differences and its "10x" finding is contested folklore. By the time the number reached the stage it had inflated a hundredfold with no supporting evidence. The Basecamp 5 experience undercuts the thesis from inside DHH's own shop. He reports the agentic approach produced an architecture "like Swiss cheese" — unreviewed, uncoordinated contributions degrading system integrity. This is the predictable failure mode of code nobody reads, and it matches Shopify CEO Tobi Lütke's recent warning about "slop grenades" in heavy agentic development. The very evidence DHH presents argues against his own conclusion. Norman's sharpest observation is structural: nothing in this keynote distinguishes a mature, stable framework from one whose creator has stopped paying attention. DHH historically drove Rails' vision. With his Ruby output at 3% and his thesis that hand-written code is economically dead, the question isn't whether Rails works today — it's who will steer it tomorrow. The Mosscap fork already argues Rails is done and needs only maintenance. Hanami has an active roadmap. Shopify contributes daily engineering. But the visionary slot is now vacant. The rhetorical strategy is pure confidence without content. When DHH finally addressed Rails developers directly, he offered "you're the best of the best" — a line Norman correctly identifies as interchangeable with any framework's conference. No roadmap, no technical direction, no acknowledgment that the community needs answers about governance and succession. The ATM analogy meant to reassure the audience about automation got its decade, economist, and teller numbers wrong — perhaps, as Norman dryly notes, a human should have double-checked. What Rails World 2026 revealed is a leadership vacuum dressed as a paradigm shift. DHH may be right that LLMs will transform software development. But using the Rails community's premier event to announce your departure from the stack, while offering nothing about who takes the wheel, is extraction of attention and legitimacy from a community that built its careers on your framework.
