YouTube Summarize
More

Rails World 2026 Opening Keynote - DHH

Ruby on Railsautoenpublicupdated

Read in about 4 minutes instead of watching 63 minutes.

AI’s inflection point

  1. DHH describes AI agents as the most exciting—and unsettling—development of his computing career because their ultimate impact remains unpredictable.
  2. He compares AI-assisted programming to photography’s democratization: cameras displaced portrait painting’s economic role but unleashed new forms of creativity.
  3. He identifies the November 24, 2025 release of Opus 4.5 as the moment capable software agents became broadly accessible.
  4. After a brief period of disappointing models, newer systems restored his confidence that frontier intelligence would keep improving and become widely available.

From writing code to directing agents

  1. DHH argues that the productivity gap between programmers using agents effectively and those working without them may now be hundreds or thousands of times.
  2. He says he produced half as much code in the past 20 months as during his previous 21 years, largely through agent assistance.
  3. At 37signals, manually writing code is now treated as an exception; when agents fail, the priority is improving the process that directs them.
  4. His central question for software organizations is now how to extract the greatest value from rapidly expanding machine intelligence.
  5. He believes English has become a more expressive programming language than Ruby because developers can specify desired outcomes directly.
  6. He characterizes his new role not as a professional programmer but as a maker who steers intelligence to create products.

Rebuilding HEY with native apps and Rust

  1. 37signals is developing six native HEY applications because agents have dramatically reduced the cost of maintaining separate platform-specific codebases.
  2. DHH predicts that many products built as web apps for economic reasons will move to native applications as development costs approach zero.
  3. He says agents make languages such as Rust attractive as implementation black boxes, providing speed and efficiency without requiring humans to write or inspect the code.
  4. The proposed Rust backend for HEY is estimated to use 99% less CPU and 95% less memory, with peak traffic potentially fitting on one Raspberry Pi.
  5. He recommends assigning agents outcomes or problems asynchronously, then reviewing completed work instead of watching tokens stream in a chat.

Rails in the agent era

  1. Rails remains well suited to web products that must work without installation, especially for temporary or low-commitment users.
  2. Convention over configuration improves token efficiency, while Rails’ one-developer philosophy aligns with agents enabling individuals to accomplish far more.
  3. Rails agent evaluations quickly reached 95% completion, requiring harder tests as model capabilities advanced.
  4. DHH advises against overly prescriptive prompts: higher-level requests and a beginner’s mindset can produce better results from capable agents.
  5. Teams will still need to determine which products belong on the web and which justify native applications.

Rethinking software architecture

  1. Agent-driven development may require reconsidering abstractions because shared abstractions can become bottlenecks when thousands of processes modify software concurrently.
  2. The traditional cost of duplication and synchronization is falling sharply, changing long-standing architectural trade-offs.
  3. Software methodologies, development cycles, responsibilities, and specifications must all be rediscovered, with no established blueprint yet.
  4. Rather than embedding separate chatbots in every product, applications should expose command-line interfaces that users’ own agents can operate.
  5. Using HEY’s CLI, an agent found an old email from vague conceptual clues that conventional keyword search could not handle.

A new era of making

  1. DHH presents Omarchy as proof that agents let makers rapidly reshape an entire computing environment around their own ideas.
  2. Omarchy installation time fell from 3 minutes 33 seconds to 35 seconds on hardware, with a nine-second result achieved in the lab.
  3. He describes rapidly creating a calculator, writing app, video trimmer, and presentation system without knowing or inspecting their C++ implementations.
  4. Agent-generated optimization can produce much smaller and faster software because machine effort makes previously uneconomical refinements affordable.

Security, uncertainty, and optimism

  1. DHH acknowledges legitimate security risks but argues that developers can use the same intelligence to build defenses, isolation, and tools such as HotCell.
  2. He warns that economic and social predictions around transformative technologies are unreliable, citing employment after ATM adoption as an example.
  3. His conclusion is that uncertainty should inspire active optimism: programmers should use their expertise to explore, compete, and help define the new industry.
  4. He urges developers to embrace the transition from hand-written code to directing agents with confidence, energy, and full commitment.

Actiepunten

  1. Adopt an asynchronous agent workflow: assign outcomes or problems, let agents work independently, and review completed results.
  2. Use higher-level prompts and avoid prescribing implementation details unnecessarily.
  3. Add a command-line interface to your application so users can operate it through their own agents; DHH challenges teams to deliver one by next Friday.
  4. Prepare for AI-related security risks and use available intelligence to build defensive tools and isolation mechanisms.