The /wayfinder Skill: Navigating the “Fog of War” of Planning
Latent Space Richard MacManus
Matt Pocock launched /wayfinder, a skill for planning when the end isn’t clear. It’s for the messy part before the tickets, when agents need a map, not just orders.
Based on reporting by Latent Space, Richard MacManus — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Matt Pocock has put out a new AI skill called /wayfinder, and it’s aimed at a problem anyone using agents runs into fast: the part where the work is still fuzzy. Pocock describes it as a way to move through the “fog of war,” when you know the project exists but can’t pin down every decision up front.
That matters because his whole pitch is about taking pressure off the planning stage. Pocock said he’d been doing a lot of overnight work with AFK agents, turning rough plans into specs and then into tickets. The part that became painful wasn’t the execution. It was keeping track of the planning itself — context windows, token counts, how deep he was going.
/wayfinder is meant to act like an orchestrator layer. Instead of one planning session doing everything, it can split work across threads, do prototyping and research, then pull the pieces back together. Pocock’s model is simple: a map for the decisions already made, a ticket for the specific job, and a session for the work the agent is actually doing.
The point of all that careful naming is not branding fluff. Pocock says agents get confused when everything is called a ticket, or when the same thing is named differently in different places. He’s been unusually focused on terminology, to the point of building an AI coding dictionary and reworking his courses around it. The idea is that good words create good handoffs.
He’s already using /wayfinder for more than engineering, including course planning and a personal website rearchitecture. And he draws a clean line between his older “grill me” skill and this one: use the former when the path is mostly visible and you just want alignment, use /wayfinder when the path itself is still hidden.
Matt Pocock isn’t selling magic here, which is refreshing. He’s selling vocabulary and process, which is a lot less sexy and a lot more useful. The AI industry keeps pretending the hard part is model capability; half the time, it’s just naming the boxes correctly so the machine stops wandering in circles.
My take — AI-written commentary, not fact-checked reporting
This is the right kind of boring. The field keeps churning out grand claims, while the real pain is still handoffs, naming, and the chaos before a plan exists. Open or closed doesn’t matter much if the agent can’t tell a map from a ticket without getting itself lost.
Read more about this at: Latent Space
Related stories
This AI entrepreneur is developing agents that can plan ahead for the unexpected
MIT Technology Review · 1 week ago ·
6