AI Is Writing the Code Now: Why the Creator of Ruby on Rails Has Put His Pencil Down

When the man who helped shape how the modern web is built says handwritten code is effectively over, it is worth pausing to ask what comes next.

AI is writing the code now, and the creator of Ruby on Rails has put his pencil down.

If you have ever used Basecamp to manage a project, checked your inbox on HEY, or visited a website without thinking about what runs underneath it, there is a reasonable chance the technology behind that experience was built with Ruby on Rails. And the man who created that framework, David Heinemeier Hansson, known across the tech world simply as DHH, has made one of the most striking statements yet about how software development is changing: his company, 37signals, has gone “pencils down” on writing code by hand.

Growing up, many of us watched science-fiction movies where characters spoke commands into supercomputers, waved their hands at holographic screens, or simply told machines what they wanted and watched the technology respond. Those scenes felt fantastical. Today, some of them feel surprisingly ordinary. We are living through the early stages of a world where people can describe what they want software to do and increasingly have AI systems build it for them.

DHH’s announcement is one of the clearest examples yet of how quickly that shift is reaching people who actually build and ship software.

Who Is DHH and Why Does His Opinion Matter?

David Heinemeier Hansson is a Danish programmer and co-founder of 37signals, the company behind productivity products including Basecamp and HEY. In the early 2000s, he created Ruby on Rails, a framework that dramatically simplified how web applications could be built and went on to become an influential part of modern web development.

Think of Ruby on Rails as a set of pre-built roads and highways that developers can use instead of having to construct every road from scratch each time they build an application.

DHH is also not simply an outside commentator observing the AI revolution. He is a programmer who has spent decades building software himself. He has also historically been skeptical of technology trends he considered overhyped. That background makes his change in approach particularly interesting.

When someone who spent much of his career writing software by hand decides that machines should now do most of that work, it is worth understanding why.

What “Pencils Down” Actually Means

At the Rails World 2026 conference in Austin, Texas, DHH delivered a keynote describing how 37signals has changed its approach to software development. The core message was straightforward: writing code manually has become the exception rather than the normal course of business at the company.

Instead of developers sitting at a keyboard and typing every line themselves, 37signals increasingly directs AI agents to carry out the implementation work. Humans remain involved in defining what needs to be built, directing the agents, reviewing what they produce and making decisions when the machines go in the wrong direction.

“Pencils down” is a phrase borrowed from examination rooms, the moment an invigilator tells students to stop writing. DHH used it to describe a similar change in software development. According to his keynote, agent-generated code has become the normal workflow at 37signals, while writing code by hand is now largely reserved for situations where an AI agent produces something that needs to be corrected.

That does not mean nobody at 37signals ever touches code manually. It means the default starting point has changed.

He Used to Be Skeptical About This

This is where the story becomes particularly interesting.

DHH did not arrive at this position as an early AI evangelist. As recently as 2025, he was openly skeptical about relying on AI for production software. He described feeling that competence was “draining out of his fingers” when he leaned too heavily on AI tools, and he kept AI assistance separate from the production code behind his products.

What changed was not simply another impressive demonstration. DHH began experimenting extensively with AI agents in real development work, using them to build features, solve problems and work on actual products.

That hands-on experience appears to have changed his assessment of what the technology was capable of doing.

The Numbers Behind the Shift

DHH did not base his argument entirely on philosophy. He also presented figures intended to show how dramatically AI-assisted development had changed his own output.

According to the figures presented in his keynote, he wrote half as much code during the previous 20 months as he had during the preceding 21 years combined. Yet in August 2026 alone, he produced about 150,000 lines of production code, compared with a historical annual average of roughly 30,000 lines.

By that measure, August alone produced roughly five times his historical annual average.

That is a striking figure, although lines of code are not a perfect measurement of software productivity. More code does not automatically mean better software, and AI-generated code can also be more verbose than code written manually. The more important point is the scale of the change DHH believes AI agents have made possible: one developer can now direct machines to perform an enormous amount of implementation work in a relatively short period.

The Mech Suit and the Kodak Brownie

To explain what working with AI agents feels like, DHH used two analogies that make the idea easier to understand even for people who have never written software.

The Mech Suit

Imagine a person stepping into a powered mechanical suit that amplifies their physical abilities. A movement that would normally require considerable effort suddenly becomes much more powerful.

That is how DHH describes AI agents. The developer remains the person making decisions, but AI systems dramatically increase how much implementation work that person can direct.

The analogy matters because it offers a different way of looking at the AI debate. Instead of asking only whether machines will replace programmers, it asks what happens when programmers are given machines capable of multiplying their output.

The Kodak Brownie

DHH’s second analogy reaches back to the history of photography.

When Kodak introduced the Brownie camera in 1900, photography became accessible to ordinary people on a scale that had previously been difficult. Portrait painting did not disappear, but photography changed its place in society. Painting could become less about producing an image because no alternative existed and more about the value of the human craft itself.

DHH draws a similar comparison with software development. His argument is not that coding disappears completely. Rather, handwritten coding could become increasingly specialized as AI systems take over more of the routine implementation work.

This Is Not Just Theory. Real Products Are Being Built This Way

One of the most concrete examples from DHH’s keynote involves HEY, 37signals’ email product.

The company is rebuilding HEY from the ground up, moving from a web application toward six separate native applications. The underlying mail server is also being rewritten in Rust, a programming language known for its performance and memory-safety characteristics.

DHH said the redesigned architecture had reduced CPU usage by 99 percent and memory consumption by 95 percent. He also suggested that a system that previously required ten servers at peak traffic could potentially run on a single Raspberry Pi.

Those are figures presented by DHH as part of his keynote, rather than independent benchmarks. Even with that qualification, they illustrate why his experience has convinced him that AI-assisted development can be used for serious production work.

A Personal Observation: I Have Lived This Shift Too

I want to pause here because this is not just a story about a Danish programmer standing on a stage in Austin. It is something many people working with technology are beginning to experience firsthand.

I have used AI to build websites, write articles, fix bugs, and handle tasks that would previously have taken me considerably longer to complete manually. The results have often been faster and less exhausting, while allowing me to spend more time thinking about the actual objective rather than wrestling with every technical step.

That is where the opportunity becomes interesting.

There is a version of this future that gives people back time, lowers the cost of building things and allows creators with strong ideas but limited technical skills to turn those ideas into working products.

But there is also an uncomfortable question sitting underneath all of this: if AI can increasingly produce functional software, what happens to the people whose profession has traditionally been built around producing that software?

The Fear Is Real. So Is the Opportunity.

The anxiety surrounding AI coding is understandable. If machines can increasingly generate, modify and debug software, developers have to ask what happens to a profession built around those activities.

But there is an important distinction between removing the need to type every line of code and removing the need for people who understand software.

DHH’s own experience at 37signals points toward that distinction. AI agents can generate enormous quantities of code, but someone still has to decide what should be built, understand the architecture, determine whether the result is actually good, identify mistakes and take responsibility when something goes wrong.

That is also where the 37signals experience provides an important warning.

When designers at the company were given the freedom to independently direct AI agents to finish features for Basecamp 5, DHH said the resulting architecture looked like “Swiss cheese”, full of holes and inconsistencies.

The lesson was not necessarily that AI agents were incapable of producing software. It was that giving machines greater freedom does not eliminate the need for experienced human oversight.

The machine can produce the code. Someone still has to understand the system.

What This Means for Developers

For developers, the skills that matter may increasingly extend beyond typing speed or memorising programming syntax.

Understanding architecture, defining problems clearly, communicating with AI systems, reviewing generated code and recognising when an apparently functional solution is actually flawed could become increasingly important.

The job title may remain “software developer”, but the day-to-day work could look very different from the traditional image of a programmer sitting alone at a keyboard for hours.

The programmer could increasingly become something closer to a director, giving instructions to multiple AI agents, checking their work, correcting their mistakes and deciding how the pieces fit together.

But that does not mean every developer will be replaced, nor does DHH’s experience prove that one person can universally substitute for an entire software team.

What it does show is that the relationship between human expertise and software production is changing.

What This Means for Everyone Else

For business owners, entrepreneurs, product managers and creative professionals, the implications could be significant.

If AI agents allow individual developers or small teams to accomplish more work with less manual coding, the cost and time required to turn certain ideas into software could fall.

That could make software development more accessible to people who previously could not afford a large technical team.

A business owner with an idea for a customer portal, an internal management system or a specialised online service may increasingly be able to work with a much smaller technical team to turn that idea into reality.

But accessibility does not remove the need for judgment. A badly designed application can still be badly designed whether its code was written by a human, an AI agent or a combination of both.

The real opportunity may therefore belong not simply to people who know how to use AI, but to people who understand what they are trying to build and can tell when the machine has built it badly.

Final Thought: The Future Is Already Running

DHH putting down his keyboard is not a eulogy for programming.

It is closer to the moment a master craftsman picks up a power tool and realises that the craft has changed without necessarily disappearing. The judgment still matters. The taste still matters. The ability to recognise good work from bad still matters.

What is changing is the amount of mechanical work a human has to perform personally.

For decades, programming meant translating human intentions into instructions that computers could execute, one line at a time. Increasingly, humans can describe what they want and allow AI systems to handle much of that translation.

The science-fiction movies were not entirely wrong.

Humans are beginning to command machines to build things at speeds that would have seemed extraordinary only a few years ago. DHH’s experience at 37signals is one particularly visible example of that transition.

AI is writing the code now.

The bigger question is what role humans will choose to play when the machines can increasingly write it.

Hot topics

How-To-Guides

Faith

Diaspora

Reviews