Matt Pocock

Matt Pocock is the creator of Total TypeScript and AI Hero, where he teaches developers to build reliable software with TypeScript and AI coding agents. His defining argument is that faster code generation makes engineering judgment more valuable: weak architecture, ambiguous requirements, and poor tests undermine automated development just as they undermine human teams.

Pocock worked as a voice and singing coach before entering software development. His early open-source contributions centered on XState, where he built tooling to improve TypeScript inference for state-machine configurations and joined the project’s core team at Stately. He later became a developer advocate at Vercel, working on Turborepo and contributing to Turbopack’s launch, before building Total TypeScript around practical instruction in generics, advanced types, and application development.

His open-source ts-reset improves TypeScript’s defaults for common JavaScript APIs. AI Hero extends the same educational mission into AI-assisted development, emphasizing the architectural and organizational decisions that determine whether coding agents produce maintainable software.

The engineering principles behind his agent workflows

• Shared understanding before implementation. Pocock’s Grill Me technique makes an agent interrogate requirements, expose hidden decisions, and establish a shared design concept before producing plans or code. He rejects specification-to-code workflows that encourage developers to ignore the resulting system; humans should retain responsibility for requirements, architecture, and product judgment.

• Vertical slices and test-driven development. He breaks features into thin, working paths across an application’s database, services, and interface instead of building each layer separately. These vertical slices expose integration failures early, while test-driven development, static checks, fresh-context reviews, and manual quality assurance provide dependable feedback. His AI-coding workflow turns clarified requirements into dependency-aware tasks that multiple agents can execute in parallel.

• Deep modules with human-owned interfaces. Pocock favors substantial modules behind small, coherent interfaces because they are easier to test, understand, and delegate. Developers design the boundaries and retain a mental model of the system; agents can implement more of the internal detail. His case for software fundamentals presents maintainable architecture as a prerequisite for effective automation.

• Small, deliberately designed agent skills. His public skills repository translates engineering practices into reusable agent instructions. He distinguishes manually invoked skills from model-selected ones, separates procedures from conditional reference material, and removes duplicated or ineffective guidance. His framework for writing agent skills also explains how consistent technical language and separate planning stages improve agent behavior.

Pocock’s sandcastle coordinates coding agents across sandboxed Git worktrees, organizing planning, implementation, review, and merging into isolated workflows. His Dictionary of AI Coding makes the field’s terminology accessible. A public remark about moving beyond local development points toward his continuing interest in how agent infrastructure could reshape everyday software development.