ASU+GSV 2025: 5 Takeaways That Are Shaping the Future of EdTech

ASU+GSV Summit and The AI Show brought together educators, developers and researchers to explore how AI can shape the future of learning. Across more than a dozen sessions, demos and conversations, our edtech team shared what we're building - and what we're learning - as we develop AI resources in collaboration with edtech developers.

At CZI, we're focused on helping developers build tools that are grounded in learning science. From balancing speed with impact to developing tools like Knowledge Graph and Evaluators, here are five takeaways that reflect how we're approaching this work - and where we see it going.

1. Research-backed tools and infrastructure make it easier to build what works.

One of the themes that came up across several sessions - including the panel "Balancing Speed With Impact," moderated by Sandra Liu Huang, CZI head of education and vice president of product - was the need for stronger connections between research and product development.

"In education, we have a responsibility to make the right thing, the easy thing," said Achievement Network CEO Michelle Odemwingie. "The right thing is research-based practices. We have a responsibility to find ways to make those the easy things in the products that we build and how we support teachers."

For developers, that means creating tools that don't just innovate quickly, but help educators bring proven practices into the classroom. At CZI, that's where we see the biggest opportunity for AI infrastructure tools - to make it easier to build with research.

What we're up to: Learning Science for EdTech: Next Steps in Our Work in Education

2. Strong AI depends on strong inputs.

To be useful in education, AI needs more than just computing power - it needs educational context and pedagogy. Frankie Warren, principal education product manager at CZI, described how developers are often building with general-purpose models that weren't designed for classrooms.

CZI's Knowledge Graph aims to change that by providing access to high-quality curriculum, state academic standards, and learning science research - all mapped together and machine-readable so developers can build with research and pedagogy first.

Without this kind of structure, even the best AI tools can struggle to generate content that's accurate, aligned or useful in a real classroom.

"We want to build a database of curriculum, standards and learning science research and aggregate that together and open it up for the public good," Warren said. "We are going to aggregate and connect this knowledge so that the AI industry and edtech can build on top of it."

Knowledge Graph gives AI systems the information they need to generate higher-quality outputs.

Warren added, "It's like giving an AI system an open book test."

Read more: Knowledge Graph Helps Build AI Rooted in Learning Science

3. Quality matters - and we need better ways to measure it.

Without rigorous evaluation, we risk innovation without impact. That challenge came up throughout the ASU+GSV Summit and The AI Show. How do we know when AI-generated content is actually good - and good for the classroom?

During a panel discussion bringing together learning scientists, practitioners and product leaders, panelists described how rubrics - long used by educators to assess the quality of student work - are becoming an essential part of evaluating AI outputs, too.

Tools like CZI's Evaluators aim to bring that kind of structured feedback into the development process, helping teams understand things like sentence complexity, vocabulary and grade-level appropriateness before content ever reaches students.

"Rubrics are the best way to use AI," Odemwingie said. "It's one of the ways in which generative AI is really strong - that they can read, analyze and compare - and with human feedback loops, get really good at it."

4. Developers need the right resources to move faster - and build smarter.

Speed matters - but so does having the right foundation to build on. Several sessions emphasized the need to support developers with the right resources - ones that make it easier to build AI-powered tools that uphold academic rigor and align with state standards.

In one session, Yusuf Ahmad, CEO of Playlab, described how early iterations of AI in education often produced inconsistent or unreliable outputs. With the right infrastructure and feedback systems, those same tools became more accurate and better aligned with classroom needs.

Resources like validation datasets and expert-backed rubrics can help developers focus on building transformative and pedagogically rigorous solutions that address the unique challenges faced by educators and schools.

5. Instructional coherence starts with better connections between tools.

Educators are navigating a fragmented edtech landscape, juggling a myriad of platforms that don't always work together, or may not be grounded in the latest learning science research and pedagogy. That lack of connection can make it harder to understand what students need - and to act on it in real-time.

What if AI could change that?

Tyler Sussman, senior education program officer at CZI, described a future where edtech tools could work together seamlessly, pulling from a rich set of pedagogical resources, aligned to state academic standards, to deliver a more precise, personalized and coherent learning experience.

Building better public AI K-12 infrastructure tools and middleware could provide the missing "glue" to more seamlessly integrate existing tools and empower teachers to truly drive instructional coherence, and focus on what matters most: student learning.

Interested in partnering with us?

We're currently in a private beta and actively partnering with developers, researchers and curriculum providers to help advance the public good of educational AI. If you're interested in learning more or getting involved, reach out to us.

Share