Fox Curriculum
AI in instructional design
An AI tool built for learning can do the thinking for the student or require it of them. The difference is a design decision, and it can be checked.
Two kinds of client, usually treated as one
Advising, design, and evaluation for two kinds of client who are usually treated as one. Each gets its own column below.
Teams building learning products with AI in them
They need to know what the product does to the learner's reasoning and not only to their output.
Who it is for
- Edtech product teams adding AI to a learning product, or building one around it.
- Publishers adding AI features to existing material.
What I do
- Review the product against the learner it is meant for. Name the cognitive work it should require, and check whether it does or whether it performs that work for the student.
- Recommend changes that are testable, written as design constraints rather than principles.
- Where the product reports gains, design the study that checks them: evidence standard set before the data arrives, question overlap tracked alongside accuracy, and the claim limited to what the design can carry.
Leaders deciding which tools their teachers and students may use
They decide which tools are allowed, under what conditions, and how anyone will know whether a tool is doing harm.
Who it is for
- District, county, and school leaders writing the guidance their teachers will run on, and vetting the tools already in their buildings, including the ones nobody approved.
- Teachers and departments who need a classroom AI policy students can follow and parents can read.
What I do
- Inventory what is in use, including what nobody approved.
- Set the gates: a student data privacy floor, an accessibility floor, and human review of any decision a tool makes about a student.
- Build the vetting instrument and run it on real tools. A gate that nothing fails is not a gate, and the results are written up with the failures included.
- Draft the guidance teachers can follow on a Tuesday and families can read, with the record of what was decided and why.
For both, the classroom-level policy if you need one. Permissive and accountable: students may use generative tools to draft or refine work and remain responsible for reviewing, editing, and approving what they submit, paired with a parent-facing disclosure of the teacher's own AI use.
Five things, each written to be checked
A written review
- A written review of the product or policy as it stands, against the learner it is for.
Testable recommendations
- Design recommendations as testable constraints, each with the check that would show it was met.
A vetting instrument
- Intake, gates, verdict values, a decision record for the file, and a plain-language summary for the people who did not sit on the committee.
Model guidance and policy
- Model guidance a district can adopt or decline, and a classroom policy with its parent-facing companion.
The study, where gains need checking
- Where reported gains need checking, you get the study design, the instruments, and a report that says what the data can and cannot support.
Instruments, policy, and one study that checked the numbers
A five-gate AI vetting instrument
Built in 2026 for a county office setting. It runs one AI tool through five gates: student data privacy (11 checks), accessibility (7), bias and consequential decisions (9), governance and procurement (9), and currency. Twelve-item intake, five verdict values, a routing table, a do-not-cite table, a watch list, and two outputs per run: a formal decision record for the file and a plain-language summary for districts. Scoped to county authority, so a failing verdict is a finding offered as service rather than a directive. In the project catalog (PDF) →
A classroom AI-use policy
In force for the 2025-26 school year. Students may use generative tools to draft or refine work and remain responsible for reviewing, editing, and approving what they submit. Paired with a parent-facing document disclosing the teacher's own AI use.
Digital practice and skill transfer
An AI-powered adaptive platform reported gains, and the room wanted to celebrate. A two-year, mentor-verified inquiry cycle set the evidence standard before the intervention began and tracked question overlap alongside accuracy. Accuracy rose from 70% to 75% across 18 students with only 5 of 30 assessment items (17%) drawn from the practice bank, which rules out memorization and leaves transfer. The study is why the gain can be trusted. Read the case study →
An ethical edtech session for school staff
Delivered with a presentation outline, speaker notes, and a written reflection, alongside a four-year technology integration progression documented across three AP History lesson plans.
The argument, made in public
The Statement on AI in Education and Work sets out the position this service is built on, and The Den carries the essays. Read the statement → Read The Den →
Where this is not the right service
- I do not build the AI model or the software. I design what it has to require of a learner and check whether it does.
- I do not give legal opinions on student data privacy law. The privacy gate is an instructional and procurement check, and where a question is legal, the instrument routes it to counsel.
- If what you want is a tool endorsement without a check, that is not a service I offer.
How it is scoped
A product review is scoped by the size of the product and the learner it is for. District and county work is scoped in phases. A first phase covers the inventory, the gates, a vetting rubric tested on tools already in use, and a written recommendation. Study design is scoped by the study.
Tell me what you are building.
Courses, assessments, a product that has to teach something. Send what you have and I will tell you what it needs.
Start a project →