DHH’s new way of writing code

The Pragmatic Engineer

0:00

-1:46:14

DHH’s new way of writing code

David Heinemeier Hansson on why he shifted to an agent-first AI workflow, and what it means for how software is built and who builds it.

Gergely Orosz

Apr 08, 2026

Stream the latest episode

Listen and watch now on YouTube, Spotify, and Apple. See the episode transcript at the top of this page, and timestamps for the episode at the bottom.

Brought to You by

• Statsig – ⁠ The unified platform for flags, analytics, experiments, and more. Stop switching between different tools, and have them all in one place.

• WorkOS – Everything you need to make your app enterprise ready. WorkOS gives you APIs to ship enterprise features in days. Check out WorkOS.com

• Sonar – The makers of SonarQube, the industry standard for automated code review. Sonar helps reduce outages, improve security, and lower risks associated with AI and agentic coding. See how SonarQube Advanced Security is empowering the Agent Centric Development Cycle (AC/DC) with new capabilities.

In this episode

David Heinemeier Hansson (DHH) is the creator of Ruby on Rails and Omarchy, co-founder and CTO of 37signals (maker of Basecamp and HEY), and the author of several books including the best-seller, Remote: Office Not Required, co-written with Jason Fried.

Six months ago, in an episode of the Lex Fridman podcast, David shared how he doesn’t use AI tools to write code: he types out all his code. But things have changed a lot since then.

In this episode, we discuss his approach to building software, how it’s changed in the last six months, and why he now takes an agent-first approach, and how he barely writes any code by hand. We go into how he uses AI agents: which alter how he builds and explores ideas, but also how his standards of quality and craft remain the same.

We also discuss how 37signals thinks about product development, from the role of designers to the importance of aesthetics and taste. David gets into how he sees beauty and functionality as closely linked, and why strong opinions about design lead to better software.

Finally, we look into the uneven impact of AI which amplifies senior engineers while creating challenges for junior developers, and what this may mean for the role of the software engineer.

Key observations from DHH

Here are 12 of my most interesting takeaways from talking with DHH:

1. His philosophy on AI has not changed, but the available tools very much have. Autocomplete-style coding assistants were genuinely annoying for experienced developers six months ago. Things changed with the shift from tab-completion to agent harnesses, plus the emergence of powerful models like Opus 4.5 – when agents started producing code which DHH does want to merge with little to no alteration.

2. Beautiful code and products aren’t matters of vanity; they’re signals of correctness. Dipping into philosophy, DHH says: “When something is beautiful, it’s likely to be correct.” He argues that Steve Jobs wanted the inside of a computer to be beautiful because people who care about circuit board layout are also those who sweat on the details of the UI.

3. DHH’s development workflow, today: he runs tmux to have two models running, and neovim in the center. Specifics:

One fast LLM running (typically Gemini 2.5) in one split terminal
A slow but more powerful model in another terminal (usually Opus)
NeoVim for reviewing diffs via Lazygit

4. Ruby on Rails seems to be enjoying a Renaissance thanks to AI. Rails is one of the most token-efficient ways of building web apps and is well-suited for agent workflows. Testing is part of the framework, which helps agents write tests and validate their own outputs. It also produces code that humans can read and verify, which matters when reviewing agent output at speed.

5. A big win from using AI agents is tackling stuff that you wouldn’t have before. A senior engineer at 37signals ran a “P1 optimization” project to improve the fastest 1% of requests. They optimized the P1 from being at 4 milliseconds, to under half a millisecond. This is the sort of work that wouldn’t have been considered previously!

6. Running several AI agents feels less like “project management” and more like “wearing a mech suit.” Being a project manager of agents did not appeal to DHH, but now that he’s building with several agents, he feels like he’s in control of the work which is being hyper-accelerated.

7. Senior engineers benefit from AI a lot more than juniors. At 37signals, senior engineers gain more from AI tools as they can validate whether an agent’s output is production-ready. DHH also notes that Amazon reached the same conclusion, and no longer lets junior programmers ship agent-generated code to production without review.

8. 37signals has one designer for every two engineers. The company has around 20 software engineers and 10 designers. Designers do far more than design; they’re also product managers and “implementers” rolled into one. On top of making things look good, they figure out what should be built, how it should work, and often build the first version. DHH compares design at 37signals to jewelry design: “you should know the properties of gold. You should know how it bends.”

9. AI agents could turn 37signals’ “designer model” into the industry standard. AI tools now empower designers to implement more of their vision directly, and DHH suspects the rest of the industry is converging toward what 37signals has always done: working with small teams, where designers are also builders.

10. Command Line Interfaces (CLI) feel like the ultimate AI interface, which validates the Unix philosophy of the 1970s. DHH is building CLIs for all 37signals products because they let agents chain tools together. “GitHub also has a CLI, and Sentry as well,” he says. “You can tie all these things together so an agent can check errors, write a fix, post a PR, and report back to basecamp.”

11. The demise of the two-month product development cycle described in the book ‘Shape Up: Stop Running in Circles and Ship Work that Matters’. The 2019 title by Ryan Signer covered how 37signals worked at the time, and DHH reveals that this methodology now needs rewriting because AI acceleration has made that timeline feel slow.

12. Eight hours of sleep is non-negotiable – even during an AI gold rush! DHH believes the dopamine loop of shipping with agents is intoxicating and can lead to higher risk of burnout. So, he sleeps eight hours and doesn’t use an alarm.