The 3-minute Centerpiece trailer · ver en español →
AI as Partner in Education
Forty minutes that set the stakes for the whole day. We tell two true stories, lay the evidence on the table — the wins and the harms — and then show, live, how to use AI the right way for a real Wenatchee classroom.
Two futures
Maria and Jake
Maria
8th grade · Wenatchee
Maria's AI tutor never gives her the answer — it asks her questions until she finds it. She wants to be an engineer at the Confluence dam, and for the first time she believes she can.
- Mastery gain
- +23%
- Critical thinking
- ↑ growing
- ✓A human teacher stays in the loop
- ✓Her school has a written, public policy
- ✓She keeps the receipts — prompts saved, work shown
Jake
same school · no policy
A chatbot wrote Jake's last seven essays. He shared an AI deepfake of his classmate Sara. No one ever taught him the rules, and no one is keeping the receipts.
- Fake-news detection
- −15.3%
- Essays written
- 7 by AI
- ✕No human checking the output
- ✕No policy, no disclosure
- ✕No receipts — nothing to verify, nothing to learn from
The difference is not talent, money, or which chatbot they used. It's a human in the loop, a written policy, and keeping the receipts. Those three habits are the whole day.
The evidence on the table
Both things are true
AI in education is producing real, measurable gains — and real, measurable harms. We're not here to sell or to scare. We're here to read the studies and decide what we do about them.
📈The wins
+23%
Mastery gain on curriculum-aligned assessments for students using Khanmigo 30+ min/week (Khan Academy data shared at ASU+GSV April 2026). Khan Academy's own June 2026 A/B results add +6.1% in next-item correctness across ~20 product tests. Khan Academy / AIMS →
+23%
Improvement for students using Khanmigo, the tutor built to ask questions rather than hand over answers. Khanmigo / Verodate →
+6.1%
Measured math gain in the Khan AIMS study — modest, durable, and biggest for the learners who struggle most. AIMS Collaboratory →
Equity-positive overall
A systematic review in npj Science of Learning finds consistent gains — strongest for ELL, special-ed, and rural students. npj Science of Learning review →
⚠️The harms
−15.3%
Drop in students' own ability to spot fake news after four weeks of over-relying on AI to do it for them (MIT). MIT / The Guardian →
1 in 3
College students regularly use GenAI on assignments; 9% admit outright cheating (Cornell, 95,000 students). Cornell, May 2026 →
Deepfakes & cyberbullying
AI-generated deepfakes are a fast-rising K-12 cyberbullying threat — exactly what happened to Jake's classmate Sara. EdTech Magazine →
The policy backlash
29 of 51 NYC Council members urged a school AI pause; the AFT called for screen bans; a Senate GAO probe is underway. NYT → · EdWeek → · Senate GAO →
Banning won't work. A free-for-all won't work. The only move left is governance — run like engineering, sized for a small town.
The live demo
Deep Think & the Model Council
The real question we'll ask, live
"I'm a 4th-grade teacher in Wenatchee with a struggling English-language-learner who's falling behind in reading. Give me a four-week, AI-partner plan — where AI helps, where it must not, and how I keep a human in the loop."
Standard Gemini
A fast, helpful first draft of the four-week plan. Useful — but it's one model's confident answer, with no visible reasoning and no second opinion. Good enough for easy questions. This one isn't easy.
Deep Think
Gemini 3 Deep Think shows its work — it reasons through the ELL student's specific needs, the FERPA limits, and what a human teacher must own. You can see why, not just what. About Deep Think →
Model Council
Three frontier AIs answer in parallel; a synthesizer reconciles them and a human decides. Where they agree is likely solid; where they disagree is where the teacher's judgment matters most. AI Council Framework →
See what three AIs would say
A mock of the Model Council on this exact question. Reveal each, then read the synthesis.
Gemini
Tap "Convene the council" to reveal.
Claude
Tap "Convene the council" to reveal.
ChatGPT
Tap "Convene the council" to reveal.
Synthesis · human decides
All three agree on the spine: AI handles the painful prep (leveled texts, vocab cards, translated parent notes) while the teacher owns every interaction with the child and keeps prompts in a Drive folder. Where they differ — how much oral practice to automate — is exactly the call a human teacher should make. That's the whole point: AI converges the easy parts so your judgment goes where it counts.
Where it all lands
The seven-point pledge
Everything in this plenary becomes seven commitments we sign — together — at the close of the day.
The Q&A panel
We've prepared for the hard questions
The 11:30 plenary closes with live Sli.do questions across five categories — positive, skeptical, hostile/policy, technical, and NCW-specific. We've drafted honest answers to all 25 in advance.
See the full question bank"Aren't you just replacing teachers?"
No. Commitment #1 is that AI is a partner, not a replacement. The teacher owns every interaction with a child. AI does the four hours of paperwork so the teacher gets back to teaching.
"What about the cheating and the deepfakes?"
Real and rising — that's Jake's story. The answer isn't a ban (bans push it underground); it's disclosure, new assessments, and keeping the receipts so misuse is visible and teachable.
"We're a small district with no budget. Where do we start?"
A spreadsheet inventory, a one-page traffic-light policy, and a parent letter. All free, all today. The Foundation's open-source tools automate the rest at no cost.