How teachers can use AI to leverage the classroom experience of their colleagues

The current wave of artificial intelligence in education is focused on one idea: AI that helps students — AI tutors; AI homework helpers; AI systems that explain concepts or generate practice questions.

Those tools can be useful. But they miss where the most important decisions in education actually happen.

The most important decision in a classroom is not what a student asks an AI system. It is what a teacher decides to do next.

Every day, teachers make hundreds of small judgment calls. Should I re-teach this concept or move on? Are my students tuning out because the work is too hard, or because they don’t see the point? Which of the dozen things I could try tomorrow is most likely to actually work?

Most of the time, teachers answer those questions the same way they always have: by drawing on their own experiences, instincts and whatever advice a colleague or coach can offer. That is not nothing, but it does mean every teacher is largely starting from scratch, learning lessons that other teachers in other classrooms have already gone through.

What if that didn’t have to be true?

The most important decision in a classroom is not what a student asks an AI system. It is what a teacher decides to do next.

At Navigator Schools, a network of public charter schools serving more than 1,800 students across California, we have been testing a different use of artificial intelligence. Instead of focusing on AI that delivers content to students, we are leveraging this technology to help us examine those hundreds of seemingly small decisions teachers must make every day.

Over years of instructional coaching, our schools have built up a large archive: thousands of classroom observations, the specific action steps teachers tried in response, and records of what happened to student performance afterward. We started asking whether AI could find patterns in that history and use them to help teachers make better decisions faster.

Consider what that looked like for Patrick Carr, an eighth grade teacher at Watsonville Prep.

For weeks, Carr was struggling to get his students to engage with a core reading skill. He tried adjusting how class started — the warm-up activity students do when they walk in the door — adding incentives and reorganizing how students worked in groups. Some things helped a little, but nothing stuck.

When our system analyzed his classroom data, it flagged a pattern it had seen before and surfaced a specific recommendation: Try a structured entry routine where students know exactly what to do the moment they sit down, pair it with short, timed practice sessions — five or six minutes of focused reading with a clear goal — and give specific students defined roles so they have a reason to stay engaged rather than drift.

None of those ideas were magic on their own. What mattered was doing them together, consistently and in a specific sequence. The system had seen that combination work in comparable classrooms before; Carr hadn’t because those classrooms weren’t his.

He tried it. Within weeks, something shifted. Students came in and got to work. The routines held and, by late winter, engagement and reading performance had both improved. There was no single breakthrough moment, but there was a clearer starting point drawn from hundreds of classrooms that had faced something similar.

That is the shift we are talking about. It’s not about replacing teacher judgment — Carr still decided how to adapt and apply every suggestion — it’s about providing teachers with a better foundation to make those judgments. 

This kind of prediction-based support is already standard in other fields. Doctors use systems that analyze patterns across thousands of patients to recommend treatments. Sports teams use historical data to call plays most likely to succeed in a given situation. Education has largely operated without anything equivalent, asking teachers to make high-stakes decisions with almost no visibility into what has worked in classrooms beyond their own.

The good news is that most schools are already sitting on the data needed to start: classroom observations, coaching notes, assessment results. What AI makes possible, for the first time at scale, is connecting that signal to a clear recommended next move, then learning from what happens when teachers try it.

This year across our network, that loop produced over 1,700 observations and 2,000 action steps — contributing to roughly a 19% improvement in instructional practice.

For decades, educators have talked about using data to improve teaching. The tools to finally make that real are here. 

The next generation of education AI will not be defined by chatbots helping students with homework. It will be defined by systems that help teachers answer the question they ask every single day: What should I try next?

•••

Daniel Whitlock is the technology innovation lead at Navigator Schools, a California charter network, and co-founder of Nova Path, an AI literacy and design studio for educators. He helps school leaders and teachers build practical AI-powered tools for instruction, assessment and operations, with a focus on fidelity, sustainability, and educator well-being.

The opinions expressed in this commentary represent those of the author.  

EdSource is dedicated to covering in California and nationally. Other publications may republish EdSource articles, online or in print, at no charge, provided adherence to our guidelines. The embed code provided here includes a tracking pixel to enable us to measure the reach of our stories. It must be included in any online pickup of EdSource stories.

The following guidelines apply to all EdSource content, including stories, podcasts and video:

Editing stories: Trimming stories for length or adding additional localized reporting is allowed. However, if you wish to make material edits, please contact EdSource for approval prior to publication. Please reach out to Andrew and Smita at republish@edsource.org with questions.

Crediting EdSource: Authors and EdSource should be credited in the byline with a link back to the original article. (This is in the embed code. Please be sure to include it.)

Monetization: EdSource stories cannot be sold, but it’s OK to put EdSource articles on pages with ads.

Photos and graphics: Assets such as photos and graphics must be added manually. Photos taken by EdSource staff may be republished, but some images, such as those from third-party providers (Associated Press, iStock), may not be republished without permission or payment. Proceed with caution. You may also run your own photos with the EdSource story. On occasion, EdSource editors may be able to provide partners with embed codes for graphics or localized data. If you have questions, please contact us.

RSS Feeds: News organizations cannot systematically republish all EdSource articles via RSS feed or other mechanism.