데이비드 하이네마이어 핸슨
Creator of Ruby on Rails, co-owner of 37signals
944 records in all
3 free questions a day, no sign-up.
Photo: David Merrett from Daventry, England CC BY 2.0 Original cropped
33 basic records of 221
Episodes are real events from public sources, summarised by us.
When the exact domain names for their products weren't available, they added extra words to create functional URLs. This allowed them to keep their preferred brand names like Backpack and Campfire.
LessonDon't let the lack of a perfect .com stop you from using a great brand name.
While building a transaction switch, the team established a rule to never process a duplicate transaction. They wrote 'Err in favor of users' on their whiteboard to guide all technical decisions.
LessonSimple, visible maxims help teams maintain consistency when building complex systems.
While developing Mach-E, David realized that his deep programming knowledge was initially a deficit because he was over-prescribing paths to the AI. He shifted to describing outcomes, finding that the agent often produced better solutions than his own prescribed methods.
LessonExpertise can sometimes lead to micromanagement; letting go of control can lead to superior technical outcomes with AI.
DHH realized that managing multiple AI agents on a single machine was too slow and disorganized. He switched to a tool called Herdr, which functions like tmux with added notifications, allowing him to manage parallel agents across multiple machines.
LessonAs your workflow evolves with AI, you must adopt or build new tools that support parallel decision-making.
David used Tailscale to connect several old mini PCs sitting in his closet across different offices in Malibu and Copenhagen. This allowed him to run 16 concurrent AI agent threads by treating remote machines as if they were local.
LessonReducing networking friction allows you to repurpose old hardware into a powerful distributed system for AI tasks.
DHH built a feature into Omarchy Quattro that monitors for application crashes and offers to let an AI agent diagnose the problem. The agent can trace errors down to specific lines of source code in external applications and offer to file detailed bug reports.
LessonAI can turn passive error logging into active, automated troubleshooting and reporting.
David tested the same coding task across multiple models including Fable, Sol, Grok, and DeepSeek. He found that while Fable was the best at planning, cheaper models like DeepSeek could complete the task at 1/20th of the cost.
LessonFounders should match the complexity of the task to the cost and reasoning power of the specific AI model.
Despite hostility toward AI in some open-source communities, Linus Torvalds publicly welcomed AI contributions to the Linux kernel. He demonstrated an open-mindedness to evolve the project with new technology despite its long history.
LessonLongevity in leadership requires the ability to look at new information and change your conclusions.
DHH tested an AI agent by asking it to sign up for his new product, Fizzy. The agent navigated the site, realized it needed an email address, and followed DHH's prompt to go to HEY.com to create one autonomously.
LessonModern AI agents can solve multi-step problems across different web interfaces using only high-level natural language directions.
DHH built a feature into his software, Omarchy, that shows what percentage of the current year and the user's expected lifespan has passed. He included this as an 'Easter egg' to remind himself and users that time is finite.
LessonVisualizing the finitude of time can serve as a powerful motivator to live more intentionally.
Basecamp decided to prioritize high-impact tasks over minor bugs. They focused on what would have the greatest impact for the greatest number of people, even if it meant leaving some bugs unaddressed.
LessonFounders should evaluate if a new feature provides more value than fixing a non-critical bug.
DHH initially agreed to change the default Git branch name from master to main to accommodate social concerns. He later realized this change achieved nothing and was merely the start of endless linguistic demands.
LessonMaking superficial changes to appease critics often fails to satisfy them and only invites further demands.
DHH observed that multi-page product specifications and business requirement documents often sat unread while the development team actually built the software. The team made progress by discussing problems and testing with users in real-time instead of following the papers.
LessonLengthy specifications are often ignored by the people actually doing the work, so focus on direct communication and building.
You are seeing 141 basic records. Membership answers from all 944 of this master's records.
See membershipSolo business and indie makers
These records are our compilation of public statements and sources. We are not affiliated with or endorsed by David Heinemeier Hansson or any related company. If something should come down, tell us. Takedown request form
