Anchor and STEER

Everything else in this curriculum is about directing AI. This page is about the point where you build something another person will use, and the questions change.

Anchor and STEER was written for people designing AI tools that teachers would use. In Build Two you are that designer. The framework was talking about you before you arrived.

Anchor comes first

A project that cannot answer these should not be built.

What professional judgment does this sharpen? Not "helps educators" but something concrete you already do and cannot do at the speed you need.

Whose workaround is this replacing, and why is that workaround inadequate? The problem is already being solved somehow, with a spreadsheet or a weekend. Your version has to be better than that, not just newer.

Does this make you more capable over time, or more dependent? A good tool is like a good coach: you start noticing things yourself. A bad one means you stop knowing your own people because the dashboard knows them.

Whose evidence have you actually heard? If you are the only voice, the base is too narrow to build on.

The Anchor card puts these on paper. It is the gate for Build Two, and it works the same way the board works for a learner.

STEER is how it behaves

Run these as a check once the thing works and before anyone else touches it.

Shared mental model. It shows its reasoning, not just its answer, and somebody using it learns something.

Teacher-set guardrails. The person using it sets the boundaries, and setting them makes them state their own practice out loud.

Escalation pathways. It knows when to stop and ask a person, and the asking is light enough to actually happen.

Evolving delegation. Trust widens with use and can always be pulled back.

Reflective feedback loops. The user learns from the exchange, not only the tool.

The distinction that carries the most weight

A tool that says "this learner is struggling with fractions" hands somebody a conclusion.

A tool that says "this learner keeps treating numerators and denominators as separate numbers, and here is a task that would tell you whether that is a concept or a notation habit" sharpens their judgment instead.

The second one is harder to build and it is the whole point.

Where this sits

Build One teaches you what it feels like to direct a collaborator. Build Two asks whether the thing you made deserves to exist and whether it treats its user as a professional.

Most people find the Anchor card harder than the build. That is the correct difficulty ordering, and it is worth saying so before anyone concludes they are bad at this.