The ASU+GSV Summit is the largest gathering in education technology. At the 2026 edition (April 12–15, San Diego), the debate had clearly shifted from “will AI transform education” to “how fast, and who gets left behind.” This report synthesizes the sessions most relevant to EdTech innovation and institutional strategy, organized by theme rather than chronology.
Summit at a glance
- 461 sessions over four days; 7,000+ attendees.
- By topic: 145 sessions tagged AI/ML, 155 K-12, 129 higher education, 83 workforce, 57 equity & access, 52 policy, 39 alternative pathways.
- Key tracks: AI & Frontier Tech, Building & Investing in EdTech, Career-Connected Learning, THE FORCE (Workforce), The Forum (Policy & Civics), Global Higher Education, and the GSV Cup.
The six signals that stood out
- AI tutoring is no longer experimental. The consensus: individualized AI tutoring becomes the default within five years. It replaces mass lecture, not human connection.
- Assessment is the next crisis. If students use AI to write, written assignments stop being reliable signals of understanding. Continuous mastery measurement and conversational (oral) assessment are the emerging answers.
- Career navigation is getting rebuilt. Continuous self-discovery, AI-powered matching, and portfolio-based evidence are replacing the “take an assessment, get three careers” model.
- AI is moving faster than institutions can adapt. Some programs rewrite curriculum every 10 weeks; traditional higher ed can’t change one in under three years.
- Bipartisan consensus on education is fragile but real. Former Secretaries of Education from both parties shared a stage; the theme was outcomes and accountability.
- AI’s effect on student cognition is the unsettled question. Platform companies and classroom practitioners were not in the same room on this.
Theme 1: AI tutoring and personalized learning
The strongest case came from Reed Hastings (Netflix founder, Anthropic board member): factories didn’t get more productive by swapping a steam engine for an electric one — productivity jumped when every machine got its own motor. Education’s “sage on the stage” is the power-distribution system; AI tutoring is the individual motor. He is funding a year-long study giving 50 random students full-time human tutors to establish the real rate of individualized learning, suspecting it is roughly 2x. Connor Zwick (Speak) reported AI language tutoring delivering fluency 3–5x faster; Alpha School compresses core academics into two hours of AI-assisted mastery learning.
The counterpoint, from Adeel Khan (MagicSchool AI): “the alchemy of wisdom and relationship creates incredible learning.” MagicSchool amplifies teachers and gives districts a toggle to turn student-facing AI off — and many use it. A recurring concern was anthropomorphism: AI models praise children far more often than humans do, and companies are incentivized to optimize for engagement. The product question for tutoring builders: what friction do you build in on purpose, and how do you measure when sycophancy is hurting the learner rather than retaining them?
For institutions: the question is no longer whether AI tutoring works — it’s whether you’re ready for students who arrive having learned 2–3x faster than your curriculum assumes.
Theme 2: The assessment crisis
If AI can write essays and take tests, how do you know what a student actually knows? Connor Zwick’s prediction was the sharpest: “Testing as a concept will feel extremely antiquated in five years,” replaced by continuous mastery measurement. Three alternatives are emerging: conversational assessment (AI-powered oral exams), continuous mastery tracking (no separate “test day”), and portfolio-based evidence. Every institution that relies on written assignments as its primary assessment needs a plan; the problem is immediate even though the solutions are early.
Theme 3: Career navigation and workforce readiness
The old model — take an assessment, get three careers, good luck — is giving way to continuous discovery and evidence of skills. Julia Dixon (ESAI) coined “narrative intelligence,” warning that volume without self-knowledge is just “scaling sameness.” The measurement gap is the recurring tension: most tools measure engagement, not whether a student got a better job. James Cryan (Willow) put a number on the credential glut — 1.1 million non-degree credentials in circulation, only about one in eight producing a wage gain. The missing layer is verification: a way for employers to trust durable-skill claims from non-elite pathways without collapsing back to the diploma as the only legible signal.
Theme 4: EdTech business and investment
The clearest founder advice came from Reed Hastings: “If you want to make money, sell to school districts, make teachers’ lives easier. If you want to change the world, focus on homeschoolers.” Sam Hyams (SpringPod) was blunt on product: “There is no excuse not to have a working prototype now,” and go straight to the user first, even when the buyer is the district. Demand is unambiguous — Coursera reported one AI-course enrollment every four seconds globally, with critical-thinking enrollments up 184% year over year.
The most useful frame came from Alex Kotran (aiEDU): “the bottleneck is no longer the technology capabilities. The bottleneck is change management.” Enterprise AI veterans echoed it: identical deployments succeed or fail on whether the organization’s people were trained and given input. ASU was cited as the counter-example — a top-down strategy paired with a bottom-up Innovation Challenge that surfaced and funded staff and faculty proposals. Most institutions, the panel argued, are missing the bottom-up half entirely.
For founders: the demand is there; the constraint is institutional readiness. The companies that win treat change management as part of the product, and they generate individual-level outcome evidence before claiming product-market fit. (That evidence problem is what ScaleU’s paid pilots exist to solve — see the pilot playbook.)
Theme 5: Education policy and reform
Two former U.S. Secretaries of Education from opposite parties shared a stage under the title “Can’t We Be Friends… It’s Not People, It’s Parties.” Rahm Emanuel called for an “education reset” tying funding to outcomes. A recurring theme among state chiefs: leaders who drive improvement often see their work dismantled by the next administration — what Hastings called “hero syndrome.” Free speech and civic education drew real concern about democratic participation.
Theme 6: Equity, access, and the compliance lever
Hastings described a one-tablet-per-child project in Rwanda; the global thesis is that software-based AI teaching, once it works, can be shared with the entire world. The sharpest market-structuring insight came from Kate Eberle Walker (Presence): special-education services are protected by IDEA, which makes delivery federally mandated. “When the guardrails of compliance are there forcing the work to happen, that’s how you get the most students served.” Mental health and wellness have no equivalent statutory backstop. The implication for founders is direct: compliance is structural budget; wellness is discretionary budget. Products that productize the compliance receipt get adopted and renew; products positioned as “wellness” win the pilot and lose the renewal.
Theme 7: AI’s effect on cognition
The most contested theme. OpenAI’s James Donovan argued the question isn’t whether AI helps cognition but how the model is tuned — tune the defaults toward pedagogy and you get metacognitive gains. The practitioner consensus was sharper and not friendly: Ben Riley (Cognitive Resonance) cited Stanford SCALE’s review of 800 LLM-in-education studies — 20 showed causal impact, virtually none positive — and named the behavior “cognitive automation,” not offloading. Larry Berger (Amplify) said every AI implementation he sees is “killing the butterfly,” the moment of collective wonder that pollinates the next thousand moments of learning.
For institutions: a tool was deployed to hundreds of millions of students before the longitudinal data exists. Wait for the RCTs and you decide with a five-year lag; deploy now without measurement and you become the data. The defensible posture, in Omar Abbosh’s (Pearson) words: “If you use it wrong, you will absolutely get dumber. If you use it right, you can get smarter.” The institution’s job is enforcing the difference.
What we’re watching next
- AI tutoring consolidates. The differentiator won’t be the tutoring — it’ll be LMS integration depth, subject specificity, and evidence of outcomes. Watch for incumbents acquiring standalone tutors.
- Conversational assessment becomes the next category, driven by the “if AI can write, how do you test?” problem.
- Student retention goes autonomous — from alerting humans toward agents that initiate interventions directly.
- Vibe-coding collapses the buyer/vendor line. Marketers, ops people, teachers, and kids now build software that used to need a dev shop; the winners wrap frontier tools with the safety and compliance layer institutions need.
- Compliance-anchored student-support vendors out-scale wellness-anchored ones — the asymmetry shows up at renewal, not the initial pilot.
If you’re building EdTech
Three takeaways run through every theme: build the human-technology bundle, not just the technology; make outcomes measurable at the individual level before claiming product-market fit; and design for zero active effort at first use. ASU ScaleU exists to help founders generate exactly that evidence — paid pilots inside Arizona State University that produce the individual-level outcomes and dollar-attributed receipts institutional buyers increasingly demand.
Apply to ASU ScaleU, read the EdTech pilot playbook, or see how to sell EdTech to universities.
Adapted from ScaleU’s ASU+GSV 2026 Summit Intelligence Report, published under CC BY 4.0. This is a knowledge resource, not an endorsement of any company or product mentioned.