Most of my career has been in or adjacent to education technology. I co-founded Interactive Purchasing Solutions and sold an e-commerce and fulfillment platform to 15 of the top 20 boarding schools in the country. I ran customer success at SchoolOne across 100+ K-12 institutions and trained 15,000+ educators. I worked on SynapticMash and ActiveProgress. I ran Declara, an AI learning platform, as CEO through its acquisition, serving the Government of Singapore, the U.S. Department of Education, state education departments in Utah and California, Australia's Ministry of Education, Mexico's national teachers' union, and universities from Stanford and MIT to Bristol and UTS. Today GLX Studio serves enterprise L&D teams, and I advise education companies and nonprofits including OurMedia, CatNClever Education, and MBZ Labs.
That is a long time watching the same problem from different seats. Here is what I have concluded.
The constraint was never distribution
Every EdTech wave promised to democratize access. MOOCs, LMS platforms, mobile learning, microlearning. Each one solved distribution and each one hit the same wall: somebody still has to make the content, and making good content is slow, expensive, and dependent on a small number of people who know the material and can teach it.
A school district has 50 curriculum specialists. A Fortune 500 L&D team has a dozen instructional designers. A nonprofit has two. The bottleneck is the same in every setting. The platform can scale to a million learners. The content team cannot.
What AI actually changes
AI does not make distribution better. It makes production cheaper by an order of magnitude. That is the first time in 20 years the binding constraint has moved.
- Source material becomes the curriculum. The policy document, the product spec, the textbook chapter, the subject-matter expert's slide deck. These can become a lesson, a video, an assessment, in minutes, without a designer touching them.
- Personalization stops being a premium feature. At Declara we built recommendation around what a learner needed next, for ministries of education and for a national teachers' union running professional development across Mexico. That was hard and expensive. Today the same intent can be expressed as "generate the version of this module for a first-year nurse in Spanish," and it is a prompt, not a project.
- Localization collapses. One asset, 75 languages, no translation vendor. For a community college system or a global workforce, that is not an improvement. It is a different model.
The platforms were ready for a decade. The content supply never was. That is what just changed.
What does not change
I am not a technology optimist by default. I have seen too many EdTech products that worked in the demo and died in the classroom. Three things AI does not fix:
- Procurement. Selling to a district, a college system, or an enterprise L&D organization is still slow, consensus-driven, and budget-cycle-bound. AI vendors who think the product will carry them through that process will learn the same lesson every EdTech founder learns.
- Trust. Educators and L&D leaders have been burned. Hallucinated content in a compliance module is a liability. At Cisco I stood up the AI governance program with Security, Legal, and Privacy: red-teaming, PII guardrails, abuse monitoring, staged rollout gates. None of that is a feature. It is the price of admission, and most education AI vendors have not paid it.
- The learner's time. Cheaper content does not create more hours in a workday or a school day. If AI only increases volume, it makes the problem worse. It has to increase relevance.
Where I think the real opportunity is
Workforce development, specifically the gap between what institutions teach and what employers need. In September 2026 I sat on the "Agentic AI in US Higher Education" panel at the California Community Colleges Future Summit, and the conversation kept returning to the same point: community colleges serve the learners with the least slack and the most to gain, and the content constraint has hit them harder than anyone. Agentic workflows that take a source document all the way to a finished, governed, multilingual learning asset are the first tool that fits their staffing reality.
If AI lets a two-person team at a community college produce what used to require a publisher, and lets an employer turn its real job requirements into learning content that the college can deliver, the pipeline from classroom to career gets shorter and cheaper for the people who most need it to. That is the use case I would build for. That is the one I am building for.
For founders in this space
Do not sell "AI for education." Pick a constraint that has blocked a specific buyer for a decade, prove you removed it, and show the number. The buyers have heard the big pitch. They want the small, provable one.