Mission Statement

We believe the great question of this age is not whether AI will transform human life, but whom it will serve. In a world increasingly shaped by systems of extraction, surveillance, dependency, and centralized control, we hold that AI must not become merely another instrument by which institutions consolidate power over individuals. It must also become a means by which individuals reclaim agency, dignity, and creative sovereignty over their own lives and work.

The Guild exists to cultivate that possibility.

We gather those who wish to master AI not as passive consumers of systems designed by others, but as skilled craftspeople capable of shaping systems worthy of human trust. We seek to develop the knowledge, practices, standards, and habits of mind required to build AI that serves human flourishing rather than human diminishment.

We affirm a simple but consequential distinction: the future should not belong solely to concierge systems that mediate life on behalf of remote institutions. It should also belong to personal systems — loyal, trustworthy, and governed in the interests of the person they serve.

For this reason, the Guild is devoted to the craft of building AI that strengthens the individual, protects the individual, and enlarges the individual's practical freedom in the world.

This is not merely a technical undertaking. It is a moral one.

It is not merely a matter of capability. It is a matter of allegiance.

And it is not merely about building better tools. It is about helping secure a more human future.

Ethics you can fail a build on. That is the standard we hold ourselves to, and it rules out most of what gets published under this heading. A commitment nobody can be caught breaking is not a commitment; it is a preference, stated warmly.

So each of the seven below is written the same way. It names something we will not do, and it names the condition under which we have failed it. If you engage the Guild and we breach one of these, you will be able to point at the sentence.

These seven elaborate the Guild's operational core, which is stated more briefly on the home page as four positions: adopt AI deliberately, reject unread output, harness the model, and disclose to the human. Where the four govern how we work, the seven govern what we owe. Where the two ever conflict, the four win.

Every system we build leaves a human able to say no and have it stick. We do not ship decision paths where the model is the last word and the person affected has no route to a human being with the authority to reverse it.

Fails when: there is no override, or the override exists but the people affected cannot find it.

When a model contributes to a decision, we can reconstruct what it saw, what it was asked, and what it returned. Prompts, model versions, and outputs are recorded like any other input to production, and retained as long as the decisions they produced remain in force.

Fails when: someone asks why the system did that, and the honest answer is that we cannot tell.

We do not put customer data into a third-party context window without a contract that covers it and consent that names it. We do not train on data collected for another purpose. We do not treat a vendor's retention policy as a substitute for our own.

Fails when: personal data crosses a boundary nobody wrote down.

We evaluate on the populations who will actually be affected, not on the median case that makes the demo work. Where a model's error rate differs across groups, we measure it, record the number, and treat an unmeasured gap as a known defect rather than an open question.

Fails when: a system ships with a single aggregate accuracy figure and nothing behind it.

We do not generate more code than the team can own. Volume produced in an afternoon becomes a liability carried for years — in review load, in token spend, in the cost of every future change. We size output to the humans who will maintain it, not to the throughput of the tool that wrote it.

Fails when: nobody on the team can explain a module that is already in production.

Our methods are open and so are our misses. When an approach we recommended did not work, we say so in public, with the specifics that would let someone else avoid it. We do not present a clean case study and quietly omit the engagement that went sideways.

Fails when: we withhold a result because it was inconvenient for us.

There is work we will not take: systems built to deceive the people they are pointed at, to launder a decision through a model so that no human has to sign it, or to manufacture consent at scale. Declining costs us revenue, which is the only thing that makes this a commitment rather than a sentiment.

Fails when: we take the engagement anyway.