Matt Pocock Debuts /wayfinder to Guide AI Agent Planning

Developer educator Matt Pocock has launched a new AI agent skill called /wayfinder, aimed at helping software engineers orchestrate complex planning phases for greenfield projects. Pocock, known for his AI Skills for Real Engineers project which has amassed over 220,000 stars on GitHub and attracted 347,000 subscribers on YouTube, designed the tool to solve the cognitive and technical limits of manual session management. When developers try to schedule extensive workloads for autonomous agents to run overnight, they often run into context window constraints and token limits.

The /wayfinder skill acts as an orchestrator layer that automatically manages planning sessions. Instead of forcing the user to constantly monitor token usage, the tool splits a large project into multiple threads. It coordinates prototyping, research, and grilling sessions, then synthesizes the results back into a centralized document. Pocock tested the workflow using Claude Code to rearchitect his personal website, demonstrating how the skill handles complex, multi-step migrations.

A core philosophy behind the tool is navigating what Pocock calls the fog of war, a state where the final project requirements cannot be decided at the outset. To resolve this ambiguity, /wayfinder establishes a ubiquitous language between the human and the AI. It uses precise, leading words to define entities like maps, sessions, and different ticket types, including research, prototype, task, and grilling tickets. This structured terminology prevents the agent from becoming confused by vague instructions.

Practitioners can choose between /wayfinder and Pocock's existing grill-me skill depending on project clarity. While grill-me is ideal for smaller features where the path forward is clear and can be aligned in a single session, /wayfinder is built for highly uncertain environments. By delegating the handoffs and context management to the orchestration layer, developers can generate highly detailed specifications and leave autonomous agents to execute large volumes of work independently.