We Build AI That Refuses to Do the Work: Ethics, Engineered Into Study Tools

The easiest thing to build in education technology is a machine that writes the paper for you. It is also the most defensible thing to refuse. We build AI that is made to prove its work — and in a domain where the customer would happily let the tool do the cheating for them, that discipline is the entire product. Here is how the same standard looks pointed at a study tool built for real college courses.

Help the student; never do the work

The line is simple to state and hard to engineer: the tool helps a student learn and never does the learning for them. The study tools are built for a student's exact course and professor, grounded in real course and professor data with a strict no-fabricated-ratings rule — if the data is not real, it is not invented. The tutor is Socratic by design: it asks the next question and surfaces the gap, rather than handing over an answer to paste. And the essay tool assesses only writing the student has already produced — built to critique a draft, instructed never to generate submittable prose. It grades; it does not ghostwrite.

Bounded by each institution's own policy

Every tool runs inside the academic-integrity policy of the specific school it serves — not a generic honor code, and not our opinion of what is fair. What counts as permitted help is the institution's call; the tool's job is to honor it. That is a deliberate transfer of authority, and it is what makes the product safe to put in front of a university.

Grounded, and checked by a different lab

None of it matters if the tool is confidently wrong, so every surface is grounded in a real corpus and checked before it ships: deterministic guardrails first — answer key, required format, a denylist of retired guidance — then an independent model, from a different lab than the one that generated the content, judging whether it holds up. The author never grades its own work. A study tool that hallucinates is worse than none, because the student cannot tell.

Where the guidance points

This is not a house opinion. UNESCO's guidance for generative AI in education is human-centered by design — protect human agency, support learning rather than replace it, validate tools before deployment — and the US Department of Education's AI in teaching and learning report lands on keeping humans in the loop. We build to that standard because the alternative — a tool that quietly makes its users worse at the thing they are paying to learn — is not a product we will ship.

The full write-up lives on jessemyers.ai.