Matt Pocock

Matt Pocock is a developer educator and the creator of Total TypeScript and AI Hero. His work makes difficult engineering practices accessible through exercises, explanations, and open-source tools—first for TypeScript developers, and increasingly for engineers directing AI coding agents.

His central argument is that faster implementation increases the value of good software design. An agent working in a confusing codebase inherits that confusion; generating more code can make the next change harder. Pocock’s approach combines requirements questioning, small working increments, testable architecture, and human responsibility for the system being built.

From voice coaching to TypeScript education

Pocock’s teaching career began with voice and singing coaching. He subsequently worked as a full-stack developer and library maintainer, becoming a member of the XState core team at Stately. His XState Codegen inspected state-machine definitions to generate types for transitions, actions, guards, and other configuration. The exploratory project was eventually superseded by type generation integrated into XState itself.

Working on XState’s types brought him into contact with TypeScript library authors and exposed the gap between everyday familiarity with the language and understanding its more powerful mechanisms. In early 2022, he began publishing short TypeScript tips. He announced plans for an advanced course while still at Stately, then took a part-time developer-advocacy role at Vercel that gave him room to develop it. His course-building partnership began with live workshops before expanding into self-paced instruction.

Total TypeScript turned that experience into an exercise-driven curriculum covering inference, generics, type transformations, and advanced library patterns. Pocock asks learners to wrestle with a concrete problem before seeing the solution, giving an abstract language feature a reason to matter. The course makes techniques used by experienced maintainers available to application developers.

His open-source ts-reset addresses smaller but recurring obstacles in the same territory. It changes the types of familiar JavaScript APIs: parsed JSON becomes unknown rather than any, requiring developers to establish what the data contains, while other adjustments make array operations more practical. The project pairs safer defaults with attention to the friction developers actually encounter.

Engineering practices for coding agents

AI Hero extends Pocock’s educational work into AI applications and agent-assisted development. Its materials range from language-model fundamentals and TypeScript AI tooling to planning, implementing, and reviewing changes. His AI Coding for Real Engineers course concentrates on bringing those practices into working engineers’ workflows.

Several ideas define his approach:

• Shared understanding before implementation. His Grill Me skill turns a loose idea into a questioning session that exposes assumptions and dependent decisions. The human must steer, disagree, and acknowledge uncertainty. Questions about how an interface feels may require a disposable prototype; prolonging the conversation cannot supply the experience of using it.

• Vertical slices and early feedback. Pocock adapts the established tracer-bullet technique to agents: build one small path through the system, test it, and expand. For a feature that reveals a video in the file system, he connected the backend operation to one interface location before adding the remaining entry points. This finds integration problems while the implementation is still small enough to understand. Test-driven development, static checks, independent reviews, and manual quality assurance give engineers feedback on what agents produce.

• Deep modules with simple interfaces. His architecture work favors substantial behavior behind a compact interface. Fragmenting a concept across many small modules can make it difficult for either an engineer or an agent to follow. Concentrating related behavior creates a clearer place to test it and lets humans retain control of design while delegating implementation.

• Small, adaptable agent skills. Pocock publishes Skills for Real Engineers as procedures developers can inspect and modify. He favors explicit control over invocation, while recognizing that this shifts responsibility onto the operator. His skill-design framework separates procedures from supporting references and loads conditional material only when needed, reducing the instructions an agent must carry through every task.

In Fixing the PR Bottleneck, Pocock applies these ideas to the gap between generating a change and making it worth reviewing. Agents make opening pull requests easier, but the surrounding engineering work still determines their quality. He calls for agent experience to receive the same rigor as developer experience: the codebase is the environment an agent works in, so improving that environment matters alongside improving the speed of code generation.

Tools for evaluation and execution

Pocock’s Evalite brings evaluation of AI applications into TypeScript. Built on Vitest, it provides a local interface for inspecting outputs, traces, and logs, with score thresholds that can fail a continuous-integration build. It makes model behavior something developers can investigate repeatedly within familiar tooling.

His Sandcastle tackles the execution side: orchestrating coding agents in isolated environments, managing branch strategies, and merging their commits. It supports parallel work and review pipelines through a TypeScript API. Evalite supplies feedback on AI application behavior; Sandcastle organizes agents’ implementation work. Both give concrete form to the engineering practices he teaches.