What is Pathfinder?

Pathfinder is a product engineering workspace for you, your team, and your coding agents. It takes each new feature from the problem to shipped code, and keeps an up-to-date picture of your product: its architecture, components, design system, and glossary. Your agent works in it from your terminal or editor, and your team works in it in the browser.

Why Pathfinder?

Agents write code faster than anyone can review it. So you face a choice: review every line and slow your agents down, or stop reviewing and lose control of your product.

Stop reviewing, and the slop builds up quickly. Unguided agents build sloppy interfaces, inconsistent copy, brittle architecture, and unmaintainable code. And slop compounds, because agents copy the code they see.

Some companies try to fix this by throwing more agents at the problem. In these "software factories", agents review each other's work in loops. That catches real bugs, and can even improve the architecture. Pathfinder uses review loops too. But an agent can only check that the product works as requested. Not whether it was the right feature to ask for in the first place.

Pathfinder takes a third path. You review the ideas your product is built from, not every line of code. The evidence behind a problem. Working prototypes built from your own components. A diagram of your architecture. A glossary your team and your agent share.

So you stay in control. You slow down for the decisions that shape your product, and your agent moves fast on the rest.

How does it work?

Each product you build is a project in Pathfinder, and each feature becomes a mission.

Every mission follows a template: a set of stages, each with a skill your agent runs. Pathfinder's default template for a product feature runs from the problem to prototypes, architecture, and the build, always in that order:

  1. Discover
    1. Understand
    2. Prioritise
    3. Specify
  2. Design
    1. Diverge
    2. Converge
    3. Prototype
  3. Deliver
    1. Architect
    2. Plan
    3. Build
    4. Verify
  4. Debrief
    1. Polish
    2. Sharpen
    3. Validate

That order is Pathfinder's opinion, and we recommend it. But your templates are yours. Change a stage, rewrite its skill, or build a template around your own way of working, and Pathfinder runs it.

Each stage ends at a decision gate. Your agent writes the stage's document with you, and nothing moves on until you approve it. If a later stage shows an earlier one got something wrong, the mission goes back to it.

Pathfinder doesn't write your code. Your agent does, in your terminal or your editor. Claude Code, Codex, Cursor, and any other MCP client connect to Pathfinder the same way. Everything you decide lands in Pathfinder's web app, where your team can read it, comment on it, and edit it.

One person pilots each mission and makes its decisions. Anyone on the team can read along, comment, or pick it up. Your agent still opens pull requests, as often as your team wants them.

What you decide stays. The problem, the prototypes, the architecture, and the words your team agreed on live in Pathfinder, and your agent updates them whenever the work changes them. So the next change starts from understanding, not from a guess at what the code does.

Your code stays in your repository, and Pathfinder never reads it. It holds only what you and your agent write into it: documents, a map of your architecture down to file names, screenshots of your components, and prototypes. None of it trains AI models.

When is it worth it?

Missions cost more time and tokens than a one-shot prompt. Most take a day or two. We know, because we build Pathfinder with Pathfinder. In return, you get understanding, quality, and alignment.

Plenty of software doesn't need to last. A quick internal tool, or a script you'll run once. If it looks rough or breaks, you make another, so one-shot it with your agent. The same goes for a quick fix to any product.

Bring Pathfinder your durable software: the product you're building to grow and last. People depend on it, so you can't just replace it, and its quality decides whether it succeeds. That's where a day or two of thinking pays you back for years.

What to read next

Pathfinder example project