TL;DR
Ruby on Rails creator David Heinemeier Hansson told the Rails World conference that 37signals, the company behind Rails, has shifted to AI agents producing nearly all of its code, declaring hand-written code over for professional work there. The claim, reported by The Pulse on Sep 24, 2026, has reignited industry debate about how quickly software engineering roles will change.
David Heinemeier Hansson (DHH), the creator of Ruby on Rails, told the Rails World conference that 37signals — the company that created and maintains Rails — has moved to AI agents generating nearly all of its code, declaring the end of writing code by hand for professional work, at least at his company. The remarks, reported by The Pulse newsletter on September 24, 2026, immediately reignited a long-running industry argument over whether manual programming is becoming obsolete.
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Speaking in his Rails World keynote, Hansson said the shift applies to 37signals’ professional work, positioning the company — long known for its opinionated, craftsman-like approach to software — as an early adopter of agent-driven development. According to The Pulse’s coverage, he framed the change as effectively settled at the company level: hand-written code is over there, even if the broader industry has not reached the same conclusion.
The report also connected the declaration to hiring signals at two of the industry’s largest employers. Amazon and Meta are struggling to hire engineers, The Pulse reported, a development that, according to the newsletter’s framing, adds weight to questions about how demand for traditional software engineering roles is shifting as AI coding tools spread.
A further prediction in the coverage: code reviews may also disappear as currently practiced. The Pulse reported the view that if agents produce most code, the human review process built around hand-written changes would likely be rethought or phased out as well. The newsletter framed the overall question as whether this change is now unstoppable across the industry, not just at 37signals.
THE PULSE · INDUSTRY DEBATE · SEPTEMBER 24, 2026
The “Death of Coding by Hand” Debate
Ruby on Rails creator David Heinemeier Hansson says AI agents now generate nearly all of 37signals’ code. His declaration puts a sharper question to the industry: how quickly will software work change?
01 / WHY THIS CLAIM RESONATES
A surprising signal from Rails
Hansson’s history of emphasizing craftsmanship makes this company-level shift a notable marker in the debate.
From craft advocate to agent-driven work
Hansson created Ruby on Rails, a major web framework, and has long been associated with software craftsmanship and skepticism toward industry fads. At Rails World, he described 37signals’ professional work as moving to code produced nearly all by AI agents.
The claim is specific to his company. It does not say every developer should stop writing code by hand, or that the broader industry has already made the same shift.
Amazon and Meta
The Pulse reported hiring difficulties at both companies, without establishing why they are occurring.
A forecast, not a finding
The report raised the possibility that agent-produced code could change or reduce today’s review practices.
What replaces review?
Teams still need ways to catch defects, share knowledge, and maintain standards.
02 / THE DEBATE
Three questions behind the headline
Agent tools have improved, but the industry still disagrees on what their spread means for engineering work.
How much can agents do?
Some executives and vendors expect large productivity gains. Working developers often report that generated code still takes human correction, review, and maintenance.
What changes for engineers?
The report has added to concerns about hiring, role expectations, and the long-term value of manual coding skills.
How will teams assure it?
If conventional code reviews change, teams will need other reliable ways to find defects and keep software maintainable.
03 / WHAT THE KEYNOTE DOESN’T PROVE
A claim is a starting point
The public report describes a company’s direction, but leaves key measurements and outcomes unanswered.
Evidence still needed
No independent data in the coverage quantifies the share of agent-generated code or measures its reliability. The phrase “nearly all” is not defined, and the effects on shipping pace, defect rates, and developer roles remain unclear.
Company statement · not an industry-wide measurement- Hiring: the reasons behind the reported Amazon and Meta difficulties are not established.
- Code reviews: their predicted decline is a forecast, not an observed industry trend.
- Transferability: results may differ across legacy systems, regulated work, and other company constraints.
- Quality: public evidence has not yet shown whether the model sustains software quality at scale.
TRACEABILITY / FROM CLAIM TO EVIDENCE
What to watch next
The debate will turn on what teams can demonstrate, measure, and repeat.
Keynote details
Review the Rails World video and slides for tools used and how agent output is validated.
Company practice
Watch for further explanation from Hansson and 37signals about how the approach works.
Jobs and adoption
Track hiring patterns and data on agent-based coding tool use across engineering teams.
Quality measures
Look for evidence on shipped products, defect rates, and whether other teams replicate the model.
THE TAKEAWAY
One company’s shift. An industry-wide question.
Hansson’s declaration is explicitly limited to professional work at 37signals. The report does not establish that manual coding is ending everywhere, or that AI agents can replace human engineering across different teams and systems.
The test ahead is evidence: measured quality, sustainable delivery, changing review practices, and results from companies beyond 37signals.
Why DHH’s Shift Carries Weight
Hansson is not a typical AI evangelist. He built Ruby on Rails, one of the most influential web frameworks of the past two decades, and has publicly championed software craftsmanship, developer happiness, and skepticism of industry fads through 37signals’ books and blog posts. His company’s move toward agent-generated code is a data point from a corner of the industry that has historically resisted automation-first narratives.
The declaration also matters because 37signals is small compared with the giants it is being discussed alongside. If a company known for deliberate, opinionated engineering can shift to agent-generated code for nearly all professional work, it strengthens the argument that the change is not limited to hyperscale firms with large AI budgets. Combined with reported hiring difficulties at Amazon and Meta, the report feeds directly into ongoing anxiety among working engineers about career prospects, skill investment, and the long-term value of manual coding ability.
The stakes extend to how software quality is maintained. If code reviews fade as a practice, one of the main mechanisms teams use to catch defects, share knowledge, and enforce standards would need to be replaced by something else — a problem the industry has not yet solved.
Products Worth Considering

Bose QuietComfort Wireless Bluetooth Noise-Cancelling Headphones, Cypress Green

Jackery HomePower 3000 Portable Power Station, 3072Wh

Samsung 32 Odyssey G5 QHD Gaming Monitor
As an affiliate, we earn on qualifying purchases.
The Long-Running Debate Over Manual Coding
The argument over whether AI will end hand-written coding has run for years, accelerating as AI coding assistants and autonomous agents improved at generating, editing, and refactoring code. Competing narratives have coexisted: vendors and some executives predict dramatic productivity gains and shrinking engineering headcounts, while many working developers report that generated code still requires substantial human correction, review, and maintenance.
37signals occupies a distinctive position in that debate. The company, co-founded by Hansson and Jason Fried, has repeatedly taken contrarian stances — including moves away from cloud dependence and public criticism of practices it considers industry hype. Hansson’s own public commentary has swung over time on AI tooling, which makes his Rails World keynote declaration a notable marker in the discussion. Rails World is the annual gathering for the Rails community, and keynotes there have historically set the tone for the framework’s direction.
The Pulse, the newsletter reporting the story, is a series from the Pragmatic Engineer covering events and trends across Big Tech and startups; its September 24, 2026 edition was devoted to the question of whether writing code by hand is over.
“The end for writing code by hand for professional work — at 37signals at least.”
— David Heinemeier Hansson (DHH), Rails World keynote, as reported by The Pulse
What the Keynote Doesn’t Prove
Hansson’s statement is a claim about one company’s practice, not a verified industry-wide measurement. No independent data was included quantifying how much of 37signals’ code is agent-generated, what quality or reliability outcomes have resulted, or how the company defines “nearly all.” It is not yet clear how the shift affects 37signals’ shipping pace, defect rates, or the roles of its human developers.
The reported hiring struggles at Amazon and Meta also lack detail in the coverage: it is not established whether they stem from AI-driven reductions in demand for traditional engineering roles, compensation issues, market conditions, or other factors. The prediction that code reviews will “probably” go away is explicitly framed as a forecast, not an observed trend. Whether agent-generated code can sustainably replace hand-written code at companies with different constraints, legacy systems, and regulatory requirements remains untested in public evidence.
Signals to Watch in the Debate
The Rails World keynote video and any accompanying slides are the natural next artifacts for engineers wanting to assess the specifics of 37signals’ approach — including which agent tools are used and how outputs are validated. Watch for follow-up commentary from Hansson, who has historically elaborated on company decisions through blog posts at 37signals.
Broader indicators to monitor include hiring and layoff patterns at major engineering employers, adoption data for agent-based coding tools, and any public reporting on whether companies that reduce code review actually maintain quality. The next round of the debate will likely turn on evidence rather than declarations: shipped products, measured defect rates, and whether teams beyond 37signals replicate the model.
Key Questions
Did DHH say all developers should stop writing code by hand?
No. His declaration was explicitly limited to professional work at 37signals — he said hand-written code is over “at 37signals at least,” according to The Pulse’s report. He did not claim the entire industry has already shifted.
Is 37signals really generating nearly all its code with AI agents?
That is the company’s own claim as stated in the keynote. No independent verification or quantified data on the share of agent-generated code, or its quality, has been published.
Are Amazon and Meta actually struggling to hire engineers?
The Pulse reported hiring difficulties at both companies, but the coverage did not establish a cause. It is unclear whether any hiring challenges relate to AI tools changing demand for engineers or to other factors.
Will code reviews really disappear?
The report framed this as a probability, not a fact. If agents produce most code, current review practices may be rethought, but no company-scale evidence of this happening has been presented.
Does this mean learning to code is pointless?
The report makes no such claim. It documents one company’s practice and an ongoing debate. What agent-heavy development means for skills, careers, and education remains unresolved.
Source: rss
