Ray Myers

Ray Myers is the founder of The Coding Agency and a former chief architect at All Hands AI, where he worked on OpenHands, an open-source coding-agent platform. His central concern is whether AI can help programmers maintain existing software without overwhelming reviewers, increasing technical debt, or weakening human judgment.

Based in St. Louis, Myers built his career around legacy systems, software reliability, testing, and programming languages. He studied at Missouri University of Science and Technology and created JHaml, a Java adaptation of the Haml templating approach. His work on refactoring and designing software for change preceded the generative-AI boom.

In 2023, he articulated his approach in the Mechanized Mending Manifesto: combine human contextual judgment, predictable algorithms, and language models to improve existing software safely. His open-source Menderbot put that idea into practice through repository analysis, documentation, code review, and type annotations.

Myers started Nopilot in 2024 to support open-source coding agents, then joined All Hands AI full time in early 2025. He subsequently founded The Coding Agency, where he explores software literacy, practical agent use, local models, and security, and co-hosts a podcast with Liz Martinez.

• Maintenance is the real test. Producing new code is easier than understanding established systems, preserving working behavior, and reducing technical debt. Myers has explored incremental refactoring informed by compiler analysis and theorem provers.

• Formal methods can constrain unreliable models. Strong typing, executable specifications, and verified refactoring can catch errors before generated changes reach human reviewers. His guide to theorem provers makes these techniques more approachable.

• Domain-specific languages offer durable abstractions. Myers favors specialized languages that express business or technical intent predictably, with AI helping create their tooling and navigate their interfaces. He questions whether natural-language prompts alone provide a stable foundation for long-lived software.

• The review economy can erase productivity gains. In his AI Engineer World’s Fair talk, Myers demonstrated an OpenHands-generated, 48-file pull request while warning that cheap code generation can simply shift the bottleneck to human review. Better specifications, automated checks, and selective prioritization matter more than maximizing output.

Myers continues publishing through his personal website and Craft vs Cruft YouTube channel. His writing on code comprehension emphasizes that reading, understanding, and taking responsibility for software remain essential even when machines can produce it quickly.