Integrity & assessment
“I can't tell anymore what my students actually know.”
AI didn’t create a bad assessment system. It revealed a design gap that was always there.
The clear-eyed read
What's actually going on
Start by letting go of the framing that this is a cheating crisis. Detection software is unreliable and falls hardest on multilingual learners; surveillance escalations poison classroom culture. The honest version of the problem is older and more interesting: our assignments used products — the essay, the worksheet, the lab report — as proxies for thinking. AI broke the proxy, not the learning.
That is the product-proxy problem, and it has a design answer. When assessment triangulates three kinds of evidence — the product, the process (drafts, decision logs, reflections a tool cannot retroactively fabricate), and performance (the student explaining, defending, applying in real time) — learning becomes visible regardless of what tools were available. The goal isn't to catch students. It's to make their thinking too visible to hide.
The practical vocabulary that gets a faculty moving: which tasks should be AI-resistant, which AI-assisted, which AI-partnered — decided on purpose, task by task, rather than by an arms race.
First moves
30 / 90 / 365 days
Sequenced, not simultaneous. The 30-day moves cost little and buy you room; the year is where the change becomes structure.
First 30 days
- Stop buying detection; start a conversation. Run one department meeting on the product-proxy problem using a real assignment as the case.
- Have each teacher run the vulnerability check on one assignment: could a student submit AI work here without my knowing? That yes is data, not failure.
By 90 days
- Pilot process evidence in a few classrooms: drafts plus a short decision log or reflection attached to major tasks.
- Add one performance layer to a major assessment — a five-minute oral defense or conference beats a detector every time.
- Agree on classroom AI-use disclosure norms so students know what use is invited, permitted, and off-limits per task.
Within the year
- Rebuild the assessment map so product, process, and performance evidence each appear in every course — the coherence work, not the pilot work.
- Align grading and reporting to the new evidence, and bring families along with plain-language communication.
Carry the work
The tools that carry this
Course 2: Redesigning Learning for an AI Future
Assignment redesign so students do the thinking even when AI is available.
Course 2A: Assessment in the AI Era
The Three-Evidence Framework, oral defense protocols, grading redesign.
"Where does AI fit in this task?" decision aid
A tailored read on any assignment: resistant, assisted, or partnered. (Coming soon.)