DHH’s new way of writing code
📚 Main topics
- From autocomplete to agentsDHH says early AI autocomplete interrupted his flow, but agent tools paired with stronger models changed his approach: he now starts projects with an agent, reviews its work, and makes changes where needed. 33:51
- A craft-centered way to buildAt 37signals, small teams of designers and developers find a product’s purpose, shape, and implementation together. DHH argues that good design and code should feel beautiful, coherent, and suited to their medium. 21:44
- More ambitious workAgents have made it practical to explore projects the team previously would not have taken on, such as optimizing the fastest 1% of requests and investigating dual-boot support for Omarchy. 1:08:06
- Changing roles and hiringDHH expects judgment, product sense, communication, and the ability to oversee complex systems to matter more as implementation gets easier. He warns that routine coding work may face pressure as developers cease to be the bottleneck. 1:16:56
✨ Key takeaways
- Senior engineers gain leverageExperienced developers can use agents in parallel and assess whether their output is sound, drawing on the architectural knowledge and judgment that juniors are still developing. 54:36
- Keep a high quality barDHH does not merge code just because an agent produced it; he expects its style and quality to meet the standards he would apply to his own work. 38:33
- Designers do more than polishAt 37signals, designers help determine what to build and how it should work, and increasingly use agents to produce functional implementations. 23:19
- Tools change the cost of explorationSince agents make prototypes and investigations cheaper, teams can test vague ideas and follow promising hunches without committing to a large project upfront. 1:10:12
🧠 Lessons learned
- Try the tools firsthandDHH suggests using a frontier model with an agent harness on a real project to understand its current capabilities, rather than trying to keep up with every new trend or tool. 41:40
- Invest in judgment and craftAs code generation becomes more accessible, the ability to choose what to build, assess its quality, and make it useful becomes increasingly valuable. 1:21:03
- Do not sacrifice health for speedDHH cautions against treating the current wave of AI as a limited-time opportunity; he recommends protecting sleep, exercise, and sustainable work habits. 1:38:10
- Keep practicing and show your workFor people seeking opportunities, DHH advises doing their best work wherever they are, continuing to learn, and putting effort into presenting their skills clearly. 1:30:20
🏁 Conclusion/next steps
- Build with agents, but stay in chargeDHH plans to keep exploring agent workflows and add CLI access to 37signals products so agents can work across tools, while people provide direction and review. 1:43:53
- Focus on valuable outcomesThe host concludes that future software professionals will need more than coding ability: they will need judgment, taste, business sense, and the ability to guide complex software toward profitable outcomes. 1:46:29
Transcript excerpt
0:00 I feel that you very much value software engineering as a craft >> hugely. I mean I think aesthetics is truth. When something is beautiful, it's likely to be correct. I think this is true in mathematics. This is true in physics. This is true in a lot of different domains. >> I wonder if there's a part of AI about the impact of doing work that we would have not done before. >> The number of projects we have tackled internally that we would never even have contemplated starting on or Legion. Jeremy, one of our most agent accelerated people, went like, "We're going to do P1. We're going to optimize
0:30 P1." Literally the fastest 1% of requests, we're going to make them even faster. There's a bit of tension right now is that most of the people I find who are all in, they're working harder than they ever have. And I've seen that with myself now, too. When you can be this effective and impactful on an hour of supervision of these agents, it's really intoxicating. And I need to go, do you know what? This is not like a limited sale. on Lex Fridman's podcast. You were still rightfully so very skeptical of AI. >> This is a nuance point and maybe it's self- serving, but I don't actually think my opinions have changed. What
🔒 The full, searchable transcript is available with Pro.
Questions & Answers
Common questions about this video
¿Qué hizo que DHH empezara a trabajar con agentes de IA como primera opción?
No dice que cambiara de opinión sobre la IA, sino que cambiaron las herramientas y sus capacidades. La combinación de agentes con acceso a herramientas como Bash y modelos capaces de producir código que él quería integrar hizo que empezara los proyectos con un agente. 37:30
¿Cómo describe DHH su flujo de trabajo actual con agentes?
Empieza las tareas con agentes, revisa los borradores y modifica el código él mismo cuando hace falta. Usa más de un modelo en paralelo y examina los cambios antes de confirmarlos. 47:19
¿Qué importancia tienen el gusto y la calidad del código para DHH cuando trabaja con IA?
Mantiene un estándar muy alto: quiere que el código de los agentes sea tan claro y pulido como el que escribiría él. No integra código descuidado solo porque funcione; espera poder revisarlo y sentirse orgulloso de incluirlo. 38:33
¿Cómo entiende DHH el papel de los diseñadores en 37signals?
Los diseñadores no se limitan a hacer que una especificación se vea bien: ayudan a decidir qué construir y cómo debe funcionar. También participan en la implementación con HTML y CSS y, a menudo, con JavaScript o Ruby. 23:19
¿Cómo puede la IA permitir que los equipos trabajen en proyectos que antes no habrían considerado?
Al bajar mucho el costo de explorar una idea, los equipos pueden probar proyectos que antes no justificaban el esfuerzo. DHH cuenta que su equipo optimizó el 1 % de las solicitudes más rápidas, reduciendo su tiempo de respuesta de unos cuatro milisegundos a menos de medio milisegundo. 1:08:06
¿Por qué cree DHH que los desarrolladores sénior pueden beneficiarse especialmente de los agentes?
Su experiencia les permite evaluar si el código generado encaja con la arquitectura y si está listo para producción, además de orientar al agente cuando se equivoca. Por eso pueden trabajar con varios agentes y multiplicar su productividad, mientras que los desarrolladores júnior podrían necesitar más supervisión. 55:07
¿Qué aconseja DHH para mantener el equilibrio al trabajar intensamente con IA?
Recomienda no tratar la IA como una oportunidad limitada que hay que aprovechar a costa de todo lo demás. Considera que dormir ocho horas, cuidar la salud y hacer ejercicio son inversiones importantes para sostener la capacidad de trabajo. 1:39:45
Related summaries
Martin Fowler & Kent Beck: Frameworks for reinventing software, again and again
The creator of Clawd: "I ship code I don't read"
VISAGENS E ASSOMBRAÇÕES - DAN PIRES, NATHAN E ROGÉRIO OLIVEIRA - Inteligência Ltda. Podcast #1978
A VERDADE SOBRE O 11 DE SETEMBRO: DANIEL LOPEZ E BORTHOLE - Inteligência Ltda. Podcast #1644
LUIS CRAZY NEAR vs 20 MULHERES ft. ASPECT, ODEMETRU & THAIZOCA - Podpaquera #42
