The GoalT Manifesto

Nothing in software exists without a reason. Not a line, not a file, not a system boundary. You can run code without knowing why it exists. You cannot govern it.

What we are not

We are not in the business of root cause. Bugs happen, tests catch them, that work is real and it is not this. GoalT has nothing to say about why something broke. It has something to say about why it exists at all.

The scale problem

Every product starts small enough for one person to hold the whole reason for it in their head. Then it grows, and it keeps growing, and at some point nobody can hold all of it anymore. This doesn't stop people from doing their jobs. It stops them from knowing why the job matters, which is a different failure, quieter, and worse.

We believe a team's long-term superiority is directly proportional to its collective fluency in its own purpose tree. Not headcount. Not process maturity. Not tooling. How many people, at any moment, could tell you why the thing they're touching needs to exist.

What that fluency buys you

A team fluent in its own purpose tree pivots:

• more robust, because effort concentrates where the graph shows real dependency, not where the loudest voice points
• more reliable, because correctness gets spent on what's load-bearing instead of spread evenly and thin
• faster, because a shift in priority prunes cleanly instead of triggering a scramble to find out what depended on what
• more maintainable, because intent survives the people who leave, instead of rotting into code nobody can explain
• with less duplicated work, because two teams solving the same purpose from different angles show up as exactly that: the same purpose, twice

Micro-focus is not fast

It looks fast. Small, well-scoped tasks close quickly, and closed tickets feel like progress. But a task closed without knowing what it serves is a decision made blind, and blind decisions get found wrong later, usually at the worst time, usually by someone else. The rework that follows is real cost, just deferred. Speed borrowed against purpose is still debt.

Competence is real, and it is a single point of failure

Ask honestly: in any project, who is actually getting things done? The answer is never "the process." It's a small number of people who understand the product well enough to make the right call without being told. That's true, and it's not comfortable, and pretending otherwise doesn't make it less true.

But we know what happens to a star topology when the center leaves: the whole network goes down with it. We don't believe in that shape. We believe in the one the internet was built on, a mesh, where purpose-fluency is distributed instead of hoarded. Competence concentrated in a few people is a liability wearing the costume of an asset. The fix isn't fewer competent people. It's more people who share what they know.

We think this changes with AI, not away from it

Implementation capacity is getting cheaper, fast, and even if that curve flattens tomorrow it has already moved further than most teams have caught up to. We're not claiming AI is worse than people at inferring purpose. We're saying the opposite is becoming true: as how gets commoditized, for what purpose becomes the interface everything else routes through, whoever, or whatever, is doing the implementing. That's our bet. Purpose is one of the few things that stays unmistakably in human hands, or at the very least, unmistakably has to be declared by someone before anything downstream can be trusted.

What this is not

This isn't a methodology to adopt wholesale, and it isn't a constraint on how you're allowed to think. It's a lens: put it on when it helps you see the shape of a project more clearly, take it off when it doesn't. If something genuinely serves more than one purpose, it gets more than one parent. If the resulting shape looks strange, that's usually because the project is strange, not because the tool is wrong.

Free, and staying that way

GoalT is Apache 2.0: open source, free to use, fork, and build on, commercial or not. It was built for people already using Claude Code, but the idea itself doesn't belong to any one tool, and we'd rather it spread than stay ours.