TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
37signals, the company behind Ruby on Rails, has moved to using agents to generate nearly all its code, according to a newsletter summary of a Rails World keynote by co-founder David Heinemeier Hansson. The change is company-specific; broader claims about coding jobs, hiring and code reviews remain unverified in the available details.
37signals, the company that created Ruby on Rails, has shifted to using agents to generate nearly all its code, according to a September 24 newsletter summary of a Rails World keynote by David Heinemeier Hansson. The announcement gives a concrete example of a software company changing its development process, while leaving open how the approach affects engineering work beyond 37signals.
Hansson said the change applies to professional work at 37signals, framing the end of hand-written code as a company-level shift. The available account does not specify which agents the company uses, how much code they generate in practice, or how the team checks their output.
The newsletter also points to claims that Amazon and Meta are struggling to hire engineers and that code reviews may eventually disappear. It provides no supporting figures or detailed evidence for either claim in the summary, so they cannot establish a broad hiring trend or forecast the end of code review.
The development is presented as part of a wider discussion about whether writing code by hand is ending. The confirmed detail is narrower: 37signals has adopted agent-generated code at large scale, as described in the keynote summary. The available information does not show that other companies have made the same change.
THE PULSE · SOFTWARE & AI · SEPTEMBER 24, 2026
The “Death of Coding by Hand” Debate
37signals says agents now generate nearly all its code. The company-level shift is concrete; what it means for software work across the industry remains an open question.
01 / WHAT THE ACCOUNT SAYS
A company-level test
The report describes a change in how one software company produces code, not a demonstrated turning point for the whole profession.
Agents at scale
Co-founder David Heinemeier Hansson said 37signals shifted to agents generating nearly all its code as part of professional work.
Tools and oversight
The summary provides no agent names, exact share of generated code, testing details, or explanation of how output is checked.
Industry-wide change
Nothing in the account shows that other companies have made the same transition or that hand-coding is ending across software development.
02 / CLAIMS VS. EVIDENCE
How far the shift extends
The newsletter places the keynote alongside claims about hiring and code reviews. Its summary does not supply evidence to verify those claims.
| Topic | What the summary reports | What remains unknown |
|---|---|---|
| 37signals workflow | ✓ REPORTED Agents generate nearly all code. | Exact share, tools, timeline, review and test process. |
| Developer work | ~ POSSIBLE Work could shift toward task definition and oversight. | Actual effects on roles, speed, quality, staffing, and maintenance. |
| Amazon and Meta hiring | ~ CLAIMED The newsletter says they are struggling to hire engineers. | Figures, sources, and evidence for a broader hiring trend. |
| Code reviews | ~ SPECULATIVE The newsletter raises the possibility they may disappear. | No evidence that 37signals ended reviews or that the practice is disappearing elsewhere. |
03 / THE OPEN QUESTION
From implementation to accountability
The practical test is how teams establish correctness, responsibility, and maintainability when agents take on more implementation work.
One reported shift.
Many unmeasured outcomes.
More agent-generated code could change what engineers do each day and what human review needs to catch. The available account does not report those effects, or say that 37signals has stopped reviewing code. Productivity and quality conclusions need results over time.
The marker reflects the scope of the reported evidence: a specific company’s stated workflow, with no demonstrated industry-wide outcome.
04 / WHAT TO WATCH NEXT
Evidence that would clarify the story
The next useful reporting would document the workflow and its outcomes, then test whether similar changes are happening elsewhere.
How agents are used
Which tools generate code, how tasks are assigned, and what human oversight looks like.
What changes over time
Measured effects on defects, delivery speed, maintenance, staffing, and code quality.
Whether the trend spreads
Evidence from other teams, plus sourcing for hiring claims and any change to code review practices.
05 / QUICK ANSWERS
Key questions
What did 37signals announce?
Hansson said at a Rails World keynote that agents now generate nearly all the company’s code. The summary gives no exact percentage or implementation details.
Will developers stop coding by hand?
No industry-wide end is established. The reported shift concerns 37signals; broader effects remain uncertain.
Are Amazon and Meta struggling to hire?
The newsletter summary makes that claim but offers no figures or supporting evidence to assess it.
Are code reviews going away?
The summary raises the possibility, but does not say 37signals ended code reviews or show that the practice is disappearing elsewhere.
What is unknown about the agents?
The account does not name the tools, explain how generated code is checked, or report effects on quality, staffing, or development speed.
What can be concluded now?
37signals has reportedly adopted agent-generated code at large scale. The account does not establish a wider shift or specify a next milestone.
A Company-Level Test of AI Coding
37signals’ decision matters because it is an identifiable change in how a software company produces code, rather than a general prediction about AI. If agents can generate most of a company’s code, that could reshape engineers’ day-to-day work toward specifying tasks, reviewing output and maintaining systems. The keynote summary does not document those effects, however, so they remain possible implications rather than reported outcomes.
The debate also bears on how companies evaluate software productivity and quality. More agent-generated code could change the amount and nature of human review needed, but the newsletter’s suggestion that code reviews may go away is not accompanied by evidence that 37signals has stopped reviewing code. For developers and managers, the practical question is how the company handles correctness, responsibility and maintenance as the tools take on more implementation work.
Rails World and the Hand-Coding Debate
The Pulse is described as a series covering events, insights and trends in Big Tech and startups. Its September 24, 2026 edition places Hansson’s Rails World keynote alongside discussion of hiring at Amazon and Meta and the future of code reviews.
Ruby on Rails is the framework associated with 37signals, making the company’s own development process relevant to software teams that use it. The summary does not provide a fuller timeline for the shift to agents or say when the change began, how it was implemented, or whether the company has published technical results.
“The end for writing code by hand for professional work — at 37signals at least.”
— David Heinemeier Hansson, as summarized by The Pulse
How Far the Shift Extends
The available summary does not identify the agents used, quantify the share of code they produce, or explain how 37signals tests and reviews generated code. It also does not report effects on staffing, delivery speed, software defects or maintenance.
Claims about Amazon and Meta hiring challenges and the possible disappearance of code reviews lack supporting detail in the provided account. It remains unclear whether they reflect specific company statements, observed data or broader commentary. No evidence here establishes that coding by hand is ending across professional software development.
Evidence Needed Beyond the Keynote
The next useful details would be a fuller account from 37signals explaining its agent workflow, human oversight and results over time. Further reporting could also clarify the basis for the hiring claims about Amazon and Meta and whether any teams are actually changing or dropping code review.
Until those details are available, the reported shift at 37signals is best treated as a company-level development in an ongoing debate. The keynote summary does not announce an industry-wide change or specify a next milestone.
Key Questions
What did 37signals announce?
David Heinemeier Hansson said at a Rails World keynote that the company had moved to agents generating nearly all its code. The available summary does not give a precise percentage or implementation details.
Does this mean developers will stop writing code by hand?
No industry-wide end is established. The reported change concerns 37signals; broader effects on professional developers remain uncertain.
Are Amazon and Meta having trouble hiring engineers?
The newsletter summary says they are struggling to hire, but the available details provide no figures or supporting evidence to assess that claim.
Are code reviews going away?
The newsletter raises that possibility, but it does not report that 37signals has ended code reviews or provide evidence that the practice is disappearing elsewhere.
What remains unknown about 37signals’ use of agents?
The summary does not name the tools, explain how generated code is checked, or report effects on quality, staffing or development speed.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
