DHH Declares the End of Manual Coding at 37signals
Ruby on Rails creator shifts to AI agents while industry quality concerns mount.
1Pencils Down at 37signals
Ruby on Rails creator David Heinemeier Hansson announced at Rails World that 37signals ceased manual coding. Writing code by hand is now treated as an exceptional error state rather than standard development. When an agent fails, engineers fix the automated workflow rather than writing code manually.
2The Agent Inflection Point
DHH compared modern AI maturity to the launch of the Kodak Brownie camera. He highlighted November 24, 2025, and the release of Opus 4.5 as the historical turning point. From that date forward, pairing with autonomous AI intelligence became economically mandatory for competitive software production.
3Architectural Shift to Rust
The company is transforming its technical architecture across multiple stacks. 37signals is migrating backend services to Rust because agents write performant Rust code reliably. Simultaneously, the team is building native mobile applications directly rather than relying purely on their traditional web application wrappers.
4Convention Simplifies AI Automation
Ruby on Rails remains the core framework for web interfaces due to its architecture. Its convention-over-configuration design pattern provides predictable structures that AI agents navigate easily. Frameworks with strict structural conventions significantly reduce generation errors when steering autonomous language models through complex feature implementations.
5Retirement From Traditional Programming
DHH revealed he retired from manual programming around March after twenty-five years. He argued that manual coding is no longer economically viable for most corporate developers. He views his new role as a professional maker steering automated intelligence rather than typing syntax into text editors.
6Rethinking Core Abstractions
Traditional software engineering heavily prioritized structural abstractions to prevent redundant code. DHH stated that abstractions lose value because the marginal cost of code repetition dropped to near zero. Engineering teams must fundamentally re-evaluate architectural patterns designed around the constraints of manual human typing.
7Expansion Across Business Functions
The adoption of AI harnesses is expanding rapidly outside pure engineering departments. Startups like Craft Docs deployed internal harnesses for non-technical staff across customer support and marketing. OpenAI shifted internal finance, recruitment, and legal operations to its automated Codex infrastructure in June 2026.
8The Factory Floor Disconnect
Widespread agent usage has introduced cultural friction inside enterprise engineering teams. An anonymous engineer highlighted widespread workplace alienation where staff work twelve-hour shifts prompting Claude Code without understanding architectures. Management pressure to maximize shipping velocity has detached engineers from thoughtful code review and system design.
9Observable Software Quality Decline
Outsourcing operational reasoning to language models has correlated with noticeable production failures. Recent incidents include unexplained platform outages on Spotify and broken feature rollouts inside Uber Eats. Teams increasingly bypass proper QA verification, mistakenly assuming that autonomous agents handle boundary conditions and localized interfaces.
10Engineering Verification Infrastructure
The core discipline of software engineering is pivoting toward automated validation systems. Organizations like Ramp and OpenAI are constructing dedicated software factories with built-in evaluation harnesses. Teams must build deterministic test environments to monitor, validate, and control non-deterministic code produced by generative models.
11Deterministic Code via LLMs
Using generative models continuously for pull request reviews remains computationally expensive and slow. A promising alternative leverages LLMs to generate strict, deterministic linter rules. Running these generated static analysis rules catches bugs faster, cheaper, and far more reliably than continuous real-time model inferences.
12The Expanding Engineering Mandate
Rather than eliminating engineering jobs, automated code generation increases architectural responsibility. Managing hallucination risks, model latency, and system safety requires deeper technical expertise. Engineers who master agent orchestration and automated verification will define software infrastructure as massive venture capital flows into autonomous tooling.
Original article: The Pragmatic Engineer — DHH Declares the End of Manual Coding at 37signals