Matt Pocock Unveils 'Teach' AI Skill

Matt Pocock, who has spent six years as a voice coach and four years teaching developers, launched a new AI 'Teach' skill to apply his comprehensive teaching philosophy. This skill aims to enable anyone to learn anything by embedding effective pedagogical principles into an AI.

Pocock successfully tested the skill by using it to learn how to solve a Rubik's Cube, demonstrating its practical application in diverse learning contexts.

Unlike stateless skills that treat each query as new, stateful skills store data, such as to the file system or MCP servers, to maintain continuity.

Pocock asserts that effective teaching inherently requires a stateful approach, as a human teacher remembers what a student has learned and what remains to be taught.

For instance, while a 'Grill me' skill might be stateless, a 'Grill with Docs' skill is stateful, tracking architectural decision records (ADRs) and glossaries to build on previous knowledge.

The 'Teach' skill is installed through the 'skills.sh' installer in the developer's skills repository. Upon activation, it first establishes a 'mission' by understanding the student's motivation and learning objectives.

Following this, the skill generates a list of high-trust primary resources from the web, forming the basis for the curriculum. This resource curation ensures that learners access credible and relevant information.

The lessons are presented as interactive HTML files, which Pocock prefers over markdown for their enhanced expressiveness and engaging format, offering a richer learning experience.

This approach aims to move beyond mere information delivery to actively encourage the development of practical skills.

The system monitors user progress and adapts subsequent lessons based on the completion of specific learning milestones, ensuring that the curriculum remains responsive to the learner's evolving needs.

By generating simple learning records from student performance, the 'Teach' skill agent mimics a real teacher to tailor future lessons. This data enables the agent to dynamically build a personalized course structure.

Like a human teacher, the agent uses this historical data to tailor subsequent lessons to the individual needs of the student, ensuring a customized and effective learning path.

The 'Teach' skill integrates links to relevant communities, encouraging learners to interact and test their ideas in real-world scenarios, a crucial step for developing deeper wisdom.

It also automatically generates a glossary for jargon and complex notation, which allows future lessons to be more concise by referencing defined terms. Additionally, cheat sheets are provided as centralized references for complex tasks.

The agent within the 'Teach' skill utilizes 'notes.md' files for internal note-taking, recording user preferences and potential issues. This system allows the agent to maintain a consistent session context across interactions.

By reviewing solve cards and progress, the agent updates its workspace state and performs diagnoses to identify gaps in a user's muscle memory or conceptual understanding.

The 'Teach' skill leverages the pedagogical concept of the 'zone of proximal development,' ensuring that learners are challenged sufficiently without being overwhelmed. The system uses the Opus 4.8 model with medium effort settings to achieve this balance.

Interactive HTML components facilitate guided, browser-based practice, allowing students to apply what they learn immediately within a practical environment.

The skill maintains a stateful context across multiple sessions by saving information to the file system, ensuring continuity and personalized adaptation throughout the learning journey.

The 'Teach' skill focuses on providing high-quality, trusted resources and interactive lessons, emphasizing the zone of proximal development to foster genuine skill acquisition. The agent is designed to encourage independent learning rather than create a permanent reliance.

The ultimate goal is to equip learners with the confidence and skills necessary to transition into real-world community engagement, effectively sending them out into the world with practical knowledge.

Onboarding new employees to a codebase remains a core application, as the system creates personalized learning paths to replace difficult-to-maintain documentation. It addresses the common challenge of outdated or hard-to-maintain traditional documentation by providing dynamic, personalized learning paths.

For instance, it can effectively teach TypeScript to a developer unfamiliar with a specific stack, tailoring the learning experience to their existing knowledge and the specific problem domain's needs.

Engineers are uniquely positioned to test AI capabilities because the current technology excels at writing code. This unique advantage allows engineers to be among the first to explore and develop new AI tools and capabilities.

The instincts and skills cultivated through this early engagement with AI are transferable to other domains beyond coding. This positions developers to shape the future of work environments, which are increasingly reliant on human-AI collaboration.

AI's proficiency in coding offers a distinct advantage, enabling developers to lead the way in integrating AI into various professional practices and innovating new solutions.

Matt Pocock encourages users to visit 'AI Hero /skills' for up-to-date information and newsletter updates on how to use his skills for engineering and other applications. He notes that skills are frequently updated, requiring users to stay current with changes, and shares personal future projects, including using the tool to learn vocal harmonies and chess openings.