In Dave Eggers’ dystopian novel The Circle, Mae Holland lands a dream job at a prestigious Bay Area tech firm. On her first day, she’s given a laptop and two monitors. She’s introduced to “the feed,” a stream of customer tickets she’s expected to resolve with a smile. It seems manageable.

Then comes the “and another thing.”

Overnight, a third monitor appears so she can network with colleagues. Then a fourth for social media presence, to prove she’s thriving. Mae isn’t forced to use them, but the implication is clear: to be disconnected is to be failing. By the end, she is drowning in a sea of glowing rectangles, her humanity partitioned into a thousand different digital tabs.

Reading The Circle made my heart race. I recognize that specific brand of anxiety because I see it every time I walk into a K-12 school. We’ve built our own “Circle,” but we call it an EdTech ecosystem.

The 3,000-Screen Problem

According to the 2025 EdTech Top 40 Report, the average U.S. school district now accesses 2,982 different tools over a single school year.

Like Mae at her desk, the modern teacher is constantly flipping. They flip from the Student Information System (SIS) to the Learning Management System (LMS); from a spreadsheet of disciplinary referrals to a notepad of handwritten parent concerns; from a literacy app to a math diagnostic.

These tools rarely speak the same language. They are islands of data. This isn’t just a technical glitch; it’s a systemic failure of coherence. We’ve bought the apps, but we’ve forgotten to build the system. This lack of interoperability is a symptom of a larger issue, incoherence.

The Human Data Router

When learning is organized around the parameters of disparate classroom experiences and apps—bell schedules, platform settings, and siloed dashboards—the student experience becomes fractured.

A student spends 20 minutes on an adaptive math platform, walks down a hallway, and logs into a literacy tool. The insights from the math tool don’t inform the literacy tool. Neither informs the LMS. And almost none of this digital footprint captures the student’s social development or emotional state.

This leaves teachers and administrators to act as human data routers. They spend their nights manually synthesizing information from a dozen dashboards, trying to infer the whole child from a collection of broken parts.

Subscribe now

When I worked as an implementation lead for early warning systems, I regularly met with teams from middle and high schools that wanted to bring about meaningful improvements in high school graduation rates. Their schools had invested heavily in data systems and tools to automate reporting about students. However, these tools and data dashboards were often useless unless the right people in the building were having conversations about the students listed in the reports. No tool could answer the question about which adult in the building had a trusting relationship with a chronically absent student. Without a clear picture of the student and their academic and social experiences in one place–with the right people in the room–the school could only approach the situation with part of the bigger picture. When I would ask the school about how they viewed data about groups of students or the entire school as a learning community, I would often learn that this never took place. The schools had never taken the approach of viewing their data from a systems level.

The Operating System Deficit

In computing, an Operating System (OS) is the layer that manages resources and allows different applications to work together. Without an OS, a computer is just a pile of expensive, disconnected hardware.

Share

K-12 education has the hardware. It has the buildings, schedules, and apps. But it has never had a true Operating System. We’ve tried to force the LMS (Canvas, Google Classroom) into that role, but those are content delivery systems—post offices, not brains.

This leads us to the critical question: Can AI become the school’s operating system?

AI as the Connective Tissue

Imagine shifting from a fragmented island model to an intelligent core. In this vision, AI isn’t just another tool on the desk; it is the orchestration layer that makes the other 2,982 tools meaningful.

• Unified Data Intelligence: Instead of 3,000 silos, an AI OS ingests data from every touchpoint—attendance, behavioral logs, extra-curriculars, and climate surveys. It creates a 360-degree, longitudinal profile. It doesn’t just report a failed quiz; it notes that the failure followed a weeklong absence that began right when the student missed the foundational lesson on ratios.

• The Universal Interface: Teachers shouldn’t have to be data scientists. They should interact with one AI agent that surfaces the most critical insights across the ecosystem: “Which students are struggling with ratios in both their homework and their science lab?”

• Automated Personalization: An AI OS can trigger actions across platforms. If a student shows mastery in one app, the OS can automatically suggest advanced, project-based learning in another, ensuring the student is always working in their Zone of Proximal Development.

The Path Forward: From Managers to Mentors

The ultimate goal of an AI OS isn’t just efficiency; It’s the restoration of the human element in education.

When a system is fragmented, humans must spend their precious cognitive energy fixing the technology. Teachers spend prep periods wrestling with spreadsheets, resetting passwords, mining emails to connect with parents, and hunting for data points. We have turned our most talented educators into administrative clerks.

Leave a comment

By implementing an intelligent OS, we offload the logistical burden of coherence to the machine. When the technology is integrated, the humans are finally free to do what machines cannot: build relationships. An AI OS can tell you that a student is struggling, but only a teacher can understand why. An AI can suggest a personalized lesson plan, but only a mentor can inspire the confidence to attempt it. The AI OS makes human connections intentional instead of sporadic.

We are currently at a crossroads. We can continue to add more screens to the desk, more apps to the tablet, and more silos to the district, effectively building our own “Circle” of burnout and incoherence. Or, we can build a new foundation. We can move past the era of the app and into the era of the system, where technology finally serves as the quiet, invisible engine that allows teachers to get back to the work of being human.

An enthusiastic math teacher from Atlanta was in a crowded room at the Colorado Convention Center this February. She was showing how she could create a game of Jeopardy for her class in a few moments. This task used to take hours. She had to look up the standards, come up with categories, questions, and rank the questions by level of difficulty. The latest version of large language models can prepare this game for her in minutes. She was beaming about how much fun this is for her and her students, and the audience was in awe. I was nodding as I thought about the possibilities.

Student engagement is a serious issue, if not a full-blown crisis, facing educators at all levels. Rates of chronic absenteeism–missing 10% or more of instructional time during the school year–nearly doubled during the pandemic and haven’t fallen back to pre-COVID levels since. Many states and districts have turned to screens as one culprit. Complete or partial mobile phone bans have been passed in many districts and states across the country. An entire movement is growing to get rid of all types of screens in schools, especially for younger learners. Lack of engagement is a common thread in the narrative around the role of technology in schools.

One way educators have been trying to better engage learners is by making the classroom experience more participatory for students. This has meant badges for completing tasks, adding points to quizzes, and doing other activities in the classroom that incorporate a sense of play. With AI, the gamification of the classroom can turn the curriculum into a responsive narrative where students participate in an elaborate academic journey. Simple activities, such as a one-off classroom Jeopardy to a semester-long epic role-playing game where the teacher serves as the campaign leader, provide new avenues to engage young minds.

Making Education Engaging, from Simple to Big Steps

Gamification is not unique to the classroom experience. Brand managers long ago figured out that adding badges and special perks keeps customers committed to their products and services. Loyalty points and special incentive programs keep customers coming back for more, driving long-term value for the company. One challenge with gamification in an educational setting is that rewards must be available to all students. Teachers who give out rewards to only a select group of students who “win” a classroom experience can undermine the educational system.

Subscribe now

Simple games, such as a one-off classroom game show, can make a review session a fun experience where students can win. Companies such as Kahoot! have taken this idea to the next level by helping teachers create simple classroom quizzes with leaderboards that track points and other features. More elaborate approaches have introduced games into other aspects of the learning environment.

One example of a classroom-based game is the Good Behavior Game, an approach to classroom behavioral management where teachers assign students to teams and deliver rewards when those teams demonstrate positive behaviors. Rigorous research shows that this model can lead to better student behaviors, teacher practices, and even academic proficiency in literacy, math, writing, and other domains. Another example is BreakoutEDU, which allows teachers to create elaborate “escape rooms” where students solve clues based on academic content.

Share

Turning from simple or even highly-structured games into immersive experiences presently means an off-the-shelf solution like Kahoot! or a scripted approach like the Good Behavior Game. Traditional educational games are static, requiring students to progress at the same pace. When I was playing Oregon Trail in the computer lab of my elementary school in the 1990s, students who perished along the trail or made it all the way to Oregon had to start over or wait for the end of the period until computer lab time was over. The game did not change for the player. With artificial intelligence, a new era of adaptive games can change the learning experience.

Resources for Gamifying the Classroom

• BreakoutEDU: https://breakoutedu.com/

• The Good Behavior Game (American Institutes for Research):

  https://goodbehaviorgame.air.org/

• Gamifying Your Class to Meet the Needs of All Learners (Edutopia): https://www.edutopia.org/article/gamifying-your-class-john-mccarthy

Teachers of the future will use AI to make their classrooms engaging, immersive, and adaptive experiences that will look more like a holodeck from Star Trek. Large language models of today are capable coders, bringing programming skills to the masses. Programs like Figma, Lovable, and Replit currently allow anyone to create their own functioning website or game. In the hands of teachers, the curriculum will be transformed into a learning environment that students can explore at their own pace. Social studies teachers will vibe code the Battle of Gettysburg and allow students to interact with Confederate and Union soldiers. A physics lesson will allow students to adjust the rules of gravity on imaginary planets. Students will be able to unlock new or supplementary content by progressing through the game. Rather than delivering content, teachers will be the guides and architects of the new classroom-based gaming experience.

Leave a comment

Ready Player One?

AI will add new tools to the classroom that will give teachers new opportunities to expand their professional roles and relationships with students. Teachers who love to design games will be able to mock up and release elaborate campaigns to their students. Putting a prompt into ChatGPT or MagicSchool today can turn a science lesson into a branching role-play quest where student decisions have a direct effect on learning progression. The teachers who push the boundaries of artificial intelligence will be able to create immersive games where students get to explore the curriculum in new and exciting ways. While this all sounds fun, the real question remains if making learning more like a game will give teachers the tools they need to better engage students.

In my last post, I took a close look at how AI can serve as an instructional coach in two ways. One way is by helping to discover effective teaching practices. The other is by recording and assessing instruction. This post picks up where that one left off by exploring two more ways: modeling and testing. Without further ado, let’s get right into it.

Modeling

Sometimes, you need to see it to believe it. When it comes to teacher professional development, teachers want to see effective instructional practices modeled in realistic classroom settings. Those opportunities are too rare for many K-12 teachers. As noted in my last post, as many as 2 in 5 public schools lack an instructional coach who can step into a classroom and model an instructional practice. While teachers have planning or prep periods, they rarely can step into another classroom to observe live instruction. This happens for a number of reasons. The schedules have to align, the school must be willing to pay for a substitute teacher, or no one in the building has the right expertise. The end result is that too few teachers have the opportunity to observe instructional practices, a key component of professional learning.

One approach that states and schools have taken is to deploy coaches to schools to strategically improve instructional practices. Mississippi has received a lot of attention lately for its dramatic improvement in elementary literacy. Between 2013 and 2024, Mississippi led the nation in gains on 4th-grade literacy based on the National Assessment of Educational Progress (NAEP), often called “the Nation’s Report Card.” This improvement has been coined “the Mississippi Miracle.” Mississippi rose from 49th in the nation to the top 10 for 4th graders. This was the result of several policy changes that required an overhaul to curriculum and instructional practices. In addition, Mississippi trained and sent reading coaches to every school in the state. That worked for early literacy and reading instruction. Providing this level of support for every subject in all grade levels would be prohibitively expensive. Fortunately, technology can scale access to professional learning.

Subscribe now

The landscape of online professional learning is robust. Platforms such as Coursera, Udemy, and EdX provide anytime access to courses on an almost unlimited number of topics. A new teacher in need of support with classroom management skills or engagement in a virtual learning environment can easily find courses on the subject with example videos. Many local libraries and CareerOneStop centers provide accounts to these services at no cost. YouTube also includes a robust library of free recorded webinars and videos. The Edutopia channel has a number of videos with classroom examples, for instance.

Resources for Online Learning

• Coursera: https://www.coursera.org/

• Edutopia YouTube Channel: https://www.youtube.com/@edutopia/videos

• EdX: https://www.edx.org/

• Udemy: https://www.udemy.com/

While self-paced, always available online learning is accessible to teachers now, there are limits in the way it can provide access to model instructional practices. Imagine a scenario where a teacher is learning about a new approach to behavioral management. The teacher wants to see this practice modeled in the classroom to help them see it in action. Unfortunately, the practice is not part of an available course they can find online. Recording a classroom is a difficult task to undertake, after all. Before recording a classroom, a waiver needs to be shared, signed by parents, and collected by the teachers and the parents of every student in the classroom. This is a huge logistical challenge for research projects I have encountered. The challenges only escalate when release forms need to be translated into multiple languages.

Now imagine a world where AI can instantly generate realistic examples of high-quality instruction and classroom interactions. AI systems are rapidly advancing in their ability to render videos. My favorite example of this is the “Will Smith eating spaghetti” videos that have gone viral in recent years. This prompt has become a de facto test for generative AI video models. The first batch produced almost terrifying images of a figure vaguely similar to Will Smith shoveling something that looked like pasta or maybe flexible PVC piping into his gaping maw. The videos reminded me more of Saturn Devouring His Son by the Spanish painter Francisco Goya than the famous actor enjoying a bowl of fresh pasta. The realism of these videos has rapidly improved since 2023. We have gone from nightmare fuel to past the uncanny valley in just a few years. Soon, if not now, AI models will be able to generate instant examples of instructional practices with realistic teacher and student interactions. Better yet, no video release forms will be required.

Share

Future AI-powered instructional coaching programs will be trained on videos of classroom interactions, infused with rich descriptions of high-quality instructional practices, and prompted by teachers seeking realistic examples of teacher-student interactions. The teachers will be able to see exemplars of pedagogy or behavior management instantly, replay it, remix it, and replicate it to their heart’s content. Where Mississippi had to find and train hundreds of reading coaches to deploy throughout the state, teachers in the future will be able to access models of instructional and classroom management practices on demand, anytime, anywhere.

Testing

A continuous improvement process cannot be completed without data to analyze. Many schools around the country rely on the Plan-Do-Study-Act model to support improvement processes. The penultimate step in the process, Study, has educators examine data to identify how to improve their instructional practices. Most educators are not trained analysts with expertise in statistics or other research methods. As I have written about frequently, they also lack the time to crunch the numbers (or participate in qualitative analysis). The process frequently derails when it comes to analyzing data.

Leave a comment

Schools can get around the problem in a number of ways, keeping things in-house or working with external partners. The internal approach can be done either by one individual teacher poring over assessment results or as a group. Lately, the groups are called professional learning communities (PLCs, since everything in education needs an acronym). In either approach, data are collected, reviewed, and analyzed to determine the effectiveness of the instructional practice being considered. This process is limited by the factors mentioned in the previous paragraph: time and skills. The external approach can address both of these constraints to a degree, as partners can often bring additional time and skills to bear. By working with a consultant or local institution of higher education, for example, a PLC can access high-quality reports on student- or school-level data.

Resources for Testing Instructional Practices

• Child Trends: A Toolkit for Educators to Explore Qualitative Data in Schools

• Regional Educational Laboratory Central: Instructional Improvement Cycle: A Teacher’s Toolkit for Collecting and Analyzing Data on Instructional Strategies

• Regional Educational Laboratory Northeast & Islands: Continuous Improvement in Education: A Toolkit for Schools and Districts

• WhatWorked Education: https://interventions.whatworked.education/

In the AI-powered future, data analysis will be a commodity that can be quickly deployed to improve teaching practices. The early versions of large language models have been less than trustworthy mathematicians. Complex, step-by-step logic has been an area of struggle for large language models (LLMs) as of early 2025. LLMs can help summarize qualitative data, however, and can be built to perform multi-step processes. The LLMs of the future will be like giving teachers their own personal Ph.D. research teams. Groups like WhatWorked Education are already creating tools to test out instructional practices at a micro scale. These applications take data with minimal formatting or cleaning and produce easy-to-understand reports about what worked (or didn’t) in the classroom.

Imagine a scenario where a PLC of social studies teachers at a high school has selected a new high-leverage teaching practice for its teachers for the year. They want to know if this practice has driven academic gains for their students. We will imagine the practice being adopted is to form and lead student discussion groups. To help them evaluate the student-led discussion groups, the PLC asks an LLM to analyze student academic performance this year and compare it to previous years when teachers had not yet been trained on this approach. Rather than spending hours mining data from a student information system, controlling for student background and other characteristics, and performing statistical analysis themselves, the LLM almost instantly produces comparison charts of student engagement and academic performance. The charts show the current year where PLC members were trying out the new instructional technique and control groups from previous years. As a result, the PLC is able to get critical feedback on how these student-led discussion groups are helping. Better yet, the tool can compare groups of students to notice differences for certain students or groups. All of this can happen in the middle of an academic year. The PLC no longer needs to clear its calendar or wait for the summer to reflect and adjust.

AI Will Help (Some) Teachers Become Scientists of Their Own Classrooms

I began this two-part post by asking if AI can serve as an instructional coach. In the real world, coaches are humans. AI systems will probably not be able to replicate that human-to-human interaction anytime soon or ever. There is something about having a human interaction that feels different and powerful. The AI-powered teacher of the future will, however, be able to thoughtfully carry out the phases of the Plan-Do-Study-Act cycle with minimal or no support from another person. Teachers will be able to identify effective instructional practices, learn how they work, see how they look in realistic classroom settings, and then gather and analyze data to assess their effectiveness. This sounds like the scientific method to me.

I like to see myself as an AI skeptic who is testing the boundaries of where and when it can work to improve teaching and learning. The power of AI, as I see it, lies in giving educators the tools to push their teaching to the highest standards. Rather than giving teachers a rigid playbook and script, the AI instructional coach will serve as a partner in discovering how, where, and when to try new approaches in the classroom. After all, a great coach can never do it for you. What they can do is help you become the best version of yourself.

The U.S. Department of Labor (DoL) released an AI Literacy Framework on Friday, February 13, 2026. Normally, this sort of framework would have come from the Department of Education (USED). With the ongoing dismantling of USED under the Trump Administration, that responsibility shifted to the DoL. The framework serves as a guide for workforce training programs and school systems across the country. One of the 7 principles in the framework caught my attention: Build Complementary Human Skills. The framework suggests that AI ought to “augment human skills, such as judgment, creativity, communication, and problem-solving.” In other words, AI should serve as a personal coach.

Many schools across the United States employ human instructional coaches. An instructional coach is someone who provides support to other teachers, typically new-on-the-job ones, about instruction. These individuals can be experienced teachers who serve in a part-time capacity while being a classroom teacher, or they are full-time employees. They do this through observing and delivering feedback, providing professional development on pedagogical or behavioral management approaches, researching effective instructional practices, collecting and analyzing data on student performance, and related support. In short, they help build complementary skills for teachers just as the AI Literacy Framework outlines.

Subscribe now

Schools used Pandemic-era funding to hire additional instructional coaches. New state laws that encourage or require the teaching of research-based literacy instructional practices further support the adoption of instructional coaches. As of 2024, 3 in 5 public school districts employed at least one instructional coach. Unfortunately, the funding for these individuals is at risk. Expanded federal education funding, such as the $190 billion Elementary and Secondary School Emergency Relief (ESSER), has ended. In the AI-enhanced world of the future, many of the core responsibilities of these individuals can now be transferred and even scaled to teachers who want or need the additional support.

Finding the Right Approach

The so-called “Science of Reading” laws being passed in state legislatures encourage or require research-based instructional approaches that support literacy. This movement, fueled by the Sold a Story podcast, recognizes the many ineffective instructional practices in school settings today. A 2023 National Center for Teacher Quality study found that as many as 2 in 5 teacher prep programs in the United States were still offering professional development on teaching practices with no evidence of improving early reading skills. When teachers and schools rely on unproven or disproven instructional practices, students suffer. One way that AI can easily support teachers is by improving access to evidence-based practices.

This is one of the main reasons why I helped create Eddie: help educators find and apply evidence-based practices in the classroom. Large language models (LLMs) can quickly read through and summarize evidence in a way that makes it accessible to practitioners. Instead of spending hours searching through individual studies, LLMs can offer up summaries of evidence-based practices on nearly any topic at any time. For teachers, an LLM is an always-available instructional coach that can deliver a quick summary and implementation tips on how the research suggests teachers should support students who struggle with the core curriculum.

Share

Where a teacher might have turned to an instructional coach in the past (assuming they had access to one in the first place) to find the right practice to support a particular student, they can now turn to an LLM anytime, day or night. They no longer have to wait for a prep period or for their instructional coach to respond to an email. The present challenge is that not every LLM is trained to support teachers with evidence in this way. Brand-name LLMs such as ChatGPT and Gemini are prone to hallucinating sources. They might base their responses on social media, blog posts, or other non-research sources that perpetuate the adoption of disproven practices. For this reason, we limited the sources Eddie can use to generate recommendations. Eddie only searches through its curated database of peer-reviewed journal articles and other evidence-based resources that the team has included. Other AI-enabled solutions take a different approach, such as MyEducationResearcher, which instantly generates a meta-analysis to combine data across studies on one particular topic.

Teachers should be careful to verify the source of the evidence when using LLMs in this capacity. One of the 5 core principles in the DoL AI Literacy Framework is to evaluate the output of AI systems. The team behind Eddie provides a direct link to the sources used to generate recommendations and instructional materials for this reason. Many currently available LLMs are helpful in accessing practitioner-friendly summaries, but they are not all at the point where they can be implicitly trusted.

Resources for Finding Evidence-based Practices

• Eddie: https://eddieai.net/

• MyEducationResearcher: https://myeducationresearcher.com/login/

The future of LLMs will help teachers find and apply evidence to support their struggling students in new ways. I see this happening in two distinct pathways.

First, LLMs will improve trust in their responses by evaluating output in real time. LLMs are more like systems that “talk” to each other. One LLM can read and edit a response before the user sees the final output. By creating a system that is trained on evaluating the output of responses, LLMs will ensure that recommendations and content (i.e., lesson plans, worksheets, assignments) are aligned with the evidence. Learning Commons is already supporting this pathway with its evaluator tools for literacy and student motivation. Future datasets from Learning Commons and other organizations will address teacher-facing applications of AI.

The other pathway will personalize outputs by tapping into data systems within schools. Schools have been drowning in a sea of data for a long time. That data rarely connects across systems, leaving a fractured picture of student experiences within a school. Student information systems serve as a warehouse of data on individual students, including demographics, course grades, and more. Unfortunately, this warehouse is filled with silos that rarely or never connect the pieces in a way that educators can leverage to diagnose issues with attendance, behavior, or academic performance. Learning management systems also contain student-level data about classroom experiences, such as assignments and learning progression. Other EdTech platforms house their own student-specific data. You can begin to see how disconnected all of these systems are for schools and individual teachers. Educators are too often in the dark about how students progress through the school system on a daily or yearly basis when these systems fail to interact with each other.

The future AI-enhanced educator will be able to quickly tap into these datasets within the school to identify trends and patterns for students. AI systems will be capable of connecting disparate databases to analyze and report on student experiences across classrooms, semesters, and years. Teachers will be able to see personalized recommendations–aligned with evidence–to support academic and behavioral outcomes for individual students, groups of students (i.e., classrooms, grade levels, English language learners), or for the entire school community. The AI instructional coach of the future will be able to instantly recognize where a student has succeeded or struggled in the past and then recommend to the teacher an appropriate intervention or support that has worked for students just like them. This will break down silos in ways that teachers and school administrators have dreamed about for ages but could never accomplish before without a team dedicated to connecting data, research, and practice within a school.

Observing and Analyzing Instruction

When I was in high school, the speech and debate teams were my life. I would spend hours after class preparing briefs for the upcoming weekend’s debate tournament. My high school’s debate coach, Jim Cavallo, was inducted into both the Indiana High School Forensic League and the National Speech and Debate Hall of Fame. “Coach C” was an early adopter and firm believer in using video recording to review and improve practice. Before an era when every teen had their own video recording device in their pocket, “Coach C” would encourage debaters to schedule and use video recording devices. There wasn’t a better way to show a young person how they repeated phrases or had verbal tics than watching it together on a videotape. This was a critical feedback tool for our team that helped Coach C amass an Indiana record 13 policy debate state champions during his renowned career.

Leave a comment

Video recording is much more accessible these days than in the 1990s when I was a high school debate nerd. Even with greater access, instructional coaching time remains scarce. A teacher could set up an iPad or phone to record a lesson. However, being able to get feedback from that recording remains a critical component of the coaching process. With more than 2 in 5 public schools lacking an instructional coach on staff, too few teachers have access to this type of feedback. Even when teachers do have instructional coaches within their building, time remains a barrier.

One of the more exciting uses of AI for teacher improvement comes from recording and analyzing instruction. AI is very good at transcribing and taking notes. By analyzing the transcripts, the latest batch of AI-powered teaching aides can share critical feedback with teachers. This feedback can show how much time a teacher spends on direct instruction, behavioral management, encouraging student dialogue, and more. Teachers can get instant analysis of their teaching style by viewing reports on the type of talking going on in the classroom. In schools where instructional coaches are unavailable or limited in their time to observe other teachers, these tools allow teachers to get the sort of feedback they need to improve instruction.

Resources for AI-powered video analysis

• ANDI Labs: https://www.andilabs.education/

• Edthena: https://www.edthena.com/

• TeachFX: https://teachfx.com/

There are clear advantages and real concerns with recording lessons. Let me address the concerns first before going fully into the tech optimist mindset. Teachers, schools, and parents are properly concerned about privacy whenever a video recording device is turned on in the classroom. Teachers might have concerns about the recording being used for evaluating their performance or ending up as fodder for social media online. Schools might be concerned about student and teacher privacy, while parents might share those same concerns for their children. Any individual teacher or school adopting these tools should have clear policies in place about how and when they will be used.

Policies that ensure that recordings will be stored in a secure server or destroyed after their use can address privacy concerns. These policies should also be shared with parents so that they can be assured that video recordings of their children will never be accessible to the public without consent. Schools and teachers should come to a shared agreement about how these videos, transcripts, and any data dashboards produced will be used. Teachers who want to track the amount of instructional dialogue, for example, may agree to share reports generated by AI-powered video analysis systems with a human instructional coach or administrator for purposes of learning and development instead of evaluation. By addressing the security and application of these technologies ahead of time with the relevant parties, schools can prevent major issues.

Now I will dive into the tech optimist mindset. The upside of these tools is quite powerful. Working with an AI-powered video analysis system, a teacher may, for example, set a goal of having students talk for at least 33% of class time (or other appropriate benchmark). They can then record a lesson once per week and track their progress towards this goal over the course of a quarter, semester, or full academic year. This kind of data would have required a person to laboriously transcribe a 45-minute class period and track the amount of time each person spoke. This was simply not feasible for resource-constrained public schools in the past. With AI, teachers have a tool to get frequent feedback on how their instruction is being delivered and received by students. To me, this is one of the most powerful ways AI can support instructional improvement in the classroom.

This is becoming a long post, so I will end it here. There are two other ways that AI will serve as an instructional coach that I want to cover in a future blog post. Let me tease this topic by suggesting that AI will help some teachers apply the scientific method in their classroom, allowing them to test and improve their practice in new ways. Stay tuned for my next post where I share how.

A recent Gallup Poll caught my attention, showing how quickly generative AI has become part of teachers' toolkits in the United States. The poll found that as many as 3 in 10 teachers used AI at least weekly during the 2024-25 school year. Those weekly users also found time savings, as much as 6 hours per week. AI is delivering on one of its core promises: increasing productivity. Another promise is more elusive. Does AI help teachers teach better?

The Gallup poll shows that AI users are reclaiming time they would have spent on administrative work. Things like drafting lesson plans, worksheets, messages to parents, and the like are being assigned to brand-name large language models like ChatGPT or education-specific tools like MagicSchool or SchoolAI. These time savings are going to other things, like creative thinking about problems affecting one student or the entire classroom. The question remains if those time savings are being redirected to better, higher-leverage activities.

At this early stage in the roll-out of AI in K-12 education, the only clear benefit is a few extra hours a week for teachers who choose to use these tools. For AI to deliver on its immense promise, it needs to prove that it leads to better educational outcomes.

In this blog, I attempt to peer into the future and predict how AI will do more than save time. The predictions start with how teachers are currently using AI before projecting the technology forward to push the envelope towards a real game-changing use case. Along the way, I provide resources that educators can use to implement AI in its current forms.

AI as a Planning Partner

The Gallup poll showed how teachers are already saving time with AI. My own EdTech product, Eddie, offers this value proposition to teachers. AI can generate a lesson plan and other instructional materials in an instant. Thanks to tools like the Knowledge Graph from Learning Commons and the CASE Network 2 from 1EdTech, large language models (LLMs) seamlessly integrate standards into generative AI tools like Eddie to generate the first draft of standards-aligned lessons. Other tools allow teachers to adapt lessons for different grade levels, adjust for shorter or longer time blocks, and more. These features are all time-savers for teachers. The next generation of AI teaching assistants holds the promise of improving the classroom experience.

Subscribe now

It is one thing to create standards-aligned instructional materials in minutes instead of hours. A greater challenge is to create instructional materials that lead directly to improved educational experiences. AI teaching assistants of the future will offer insights into improving the lesson by drawing on effective practices. Instead of suggesting a generic small group discussion, the lesson plans will include activities based on learning science and tailored to the particular challenges facing the classroom and even individual students. AI systems can instantly read through large databases to filter for discussion activities that have proven to be effective for the classroom setting and/or individual student. This is the second value proposition behind Eddie, helping educators infuse evidence-based practices into classroom activities. A high-quality teaching assistant will act like an always-available instructional coach, offering guidance based on the evidence, aligned to the needs of individual students.

Resources for Lesson Planning

• TeachAI planning guides: https://www.teachai.org

Other time-saving benefits of AI teaching assistants include differentiation, scaffolding, and accessibility. These systems can already rewrite texts at different reading levels or create scaffolds for English learning and students with individualized education plans (IEPs). Imagine a 5th-grade teacher with a student who has an undiagnosed learning disability. Instead of waiting for a diagnosis, the teacher has their AI-powered teaching assistant relevel a reading assignment for that student. This happens in a moment, allowing the teacher to support the student right there in the classroom. Additional benefits will include accessibility recommendations aligned with principles of Universal Design for Learning. Scaffolding. With AI teaching assistants, teachers will be able to provide the kind of individual or small group instruction necessary to support students in the moment.

Resources for Differentiation, Scaffolding, and Accessibility

• Diffit (leveled texts & scaffolds): https://www.diffit.me

• CAST UDL resources: https://www.cast.org

AI as a Feedback and Assessment Assistant

Few parts of teaching matter more than feedback. Unfortunately, few are more time-consuming. Thoughtful, timely feedback helps students understand what they’re doing well, where they need to improve, and what to do next. Yet in many classrooms, feedback is delayed or abbreviated simply because there isn’t enough time in the day. AI can help by taking on some of the first-pass work of assessment, allowing teachers to focus on instructional judgment and student relationships.

Share

AI teaching assistants can boost the use of formative assessment. Formative assessments such as exit tickets, checks for understanding, and practice questions aligned to a lesson’s objective can all be generated by AI. Instead of starting from a blank page, teachers can ask AI to draft multiple-choice questions, short-answer prompts, or discussion questions at varying levels of rigor. While AI is doing the initial draft, teachers remain the human in the loop, serving as the editor and final decision-maker to review, refine, and select questions that make sense for their students.

Providing individualized feedback is one of the highest-impact instructional practices. However, implementing this practice for every student can be overwhelming, especially when teachers are saddled with large class sizes. AI can help by drafting rubric-aligned feedback based on student work or common response patterns. For example, a teacher might use AI to generate feedback comments tied to specific rubric criteria, draft suggestions for revision or next steps, or identify strengths and misconceptions across a set of responses. Creating the first draft of these activities is time-consuming. When the AI teaching assistant takes the first pass, teachers remain in the driver’s seat and edit feedback before students ever see it. Used this way, AI supports consistency and efficiency without replacing professional judgement.

Resources for Assessments

• Edutopia on assessment practices:https://www.edutopia.org/assessment

• Edutopia on delivering effective feedback: https://www.edutopia.org/article/13-tips-to-quickly-deliver-super-effective-feedback

The future of AI teaching assistants will allow teachers to deploy individual AI agents to students, allowing each student the opportunity to instantly gather feedback based on the teacher’s rubric. In the middle of drafting a written prompt, a student will be able to ask the AI teaching assistant how their draft aligns with the assignment and where they can improve. The teacher will be able to simulate individual feedback to a whole class, freeing the teacher to focus on students who need the most support. Checks for understanding will be deployed in real time to each student with summaries delivered back to the teacher, allowing them to pause a lesson and address misconceptions shared by the whole class. Teachers who take advantage of these features will be able to scale their feedback in new ways that push students to perform at higher levels.

AI as Connecter

The status quo for schools is a sea of disconnected technology. An Instructure report from June 2025 estimates that the average school district in the United States has 3,000 distinct EdTech tools. When these tools fail to communicate with each other, learning happens in isolation. Unfortunately, the proliferation of AI tools is only making matters worse for the time being.

Leave a comment

Future AI teaching assistants will be able to piece together the sea of systems into a stream of insight. Data housed in student information systems (warehouses of data about students, such as demographics, grades, and behavioral incidents), learning management systems (platforms to manage individual classroom experiences, including lessons, assignments, and assessments), and other EdTech programs will come together to paint a fuller picture of student learning. When a teacher needs to support an individual student, the AI teaching assistants of the future will automatically draw on other EdTech platforms to identify patterns and trends to suggest appropriate supports to the teacher. For example, the AI teaching assistant will know when a student thrived when a Socratic tutoring bot was assigned by another teacher or when a daily check-in from a trusted adult helped them avoid disruptive classroom outbursts. The AI teaching assistant will review students’ profiles to recommend evidence-based practices that have been effective for similar students. When spontaneous hallway conversations in the past sometimes led to shared insights between teachers, AI teaching assistants in the future will bridge the communication gap with interoperable systems.

Better Teaching Is Still Human

Education is built on strong relationships. AI cannot currently replace relationships, empathy, or expertise that teachers provide. I do not believe it will be able to do that within 10 years, either. What AI can and will be able to do is make life easier for teachers, saving their precious time. Future AI tools will be better able to support teachers by being a thought partner in the classroom that can also communicate with other databases to deliver helpful and timely recommendations. The AI-enhanced teachers of the future will have a guide that helps them build stronger relationships with the students who need their attention the most.

Teachers can begin to push the boundaries of AI in the present by exploring how these tools can make their work experiences easier and more effective. Begin by exploring one task that AI can support, using one tool, to accomplish one goal. Start with a quick win that can be repeated throughout the school year. As AI tools become more useful, early adopters will be poised to take full advantage of new features. The AI-powered teachers will still be making the decisions, but faster and better than before.

Conservative education group the Fordham Institute published a report on the perils of grade inflation in 2018, shortly before I wrote the first draft of this blog. The topic is a big deal to the think tank; They wrote about it again in 2023 and 2024. Grade inflation is a phenomenon where grades for students gradually increase over time. The implication is that it is easier to get a good grade as teachers relax their standards. In the original 2018 article, the Fordham Institute tapped American University economist Seth Gershenson to explore 10 years of high school transcripts and grades in Algebra 1, a required course for all students, for all public school students in the state of North Carolina.

In a follow-up blog post featuring the Gershenson study for the foundation’s website, Fordham’s Brandon Wright recounted a series produced by the Boston Globe of valedictorians who graduated from high schools in the city. One such valedictorian struggled in college at Bryn Mawr College, a highly selective private women’s college in Pennsylvania. The culprit, according to Wright, was that inner-city high schools were handing out easy A’s to students who had not earned the grades.

Subscribe now

The proof offered by Wright in the Fordham blog post is the study of North Carolina high schools by Gershenson. Curiously, Gershenson concluded that grade inflation was occurring more often at schools filled with well-off students rather than in the underfunded inner-city or rural schools in the state. Over the 10-year period studied, Gershenson found that the median grade in schools with mostly students from well-off families rose by a little less than one-third of a letter grade (the difference between an A- and an A) while those with predominantly less well-off students rose by about half of that. Kevin Mahnken, writing for education blog The 74 came to the opposite conclusion of Wright after reading the Gershenson study, noting that wealthy schools were more apt to suffer from grade inflation at the hands of parents eager to boost their students’ grades.

The claims of both Gershenson and Wright deserve attention. Will a steady rise in the average letter grade given to students lessen the ability to rely on grades as a measure of academic success in school, as well as an accurate predictor of future success in college? First, we need to take a closer look at what is going on with grade inflation.

In the Gershenson study, schools in North Carolina experienced rising high school graduation rates over the decade of data analyzed. This tracked with the national trend as high schools across the country saw an increase in graduation rates leading up to the Pandemic. College entrance exam scores in North Carolina and around the nation were more or less flat at the same. This leads Gershenson to conclude that grades were going up, but other measures suggested students were ill-prepared for postsecondary education. Gershenson concluded that, because a significant number of students who earned a B in Algebra 1 scored below proficient on the state’s end-of-course examination, a required standardized assessment administered to all students completing Algebra 1 in the state, many students were not prepared for the demands of a college course.

Share

While high school graduation rates rose in North Carolina during the time of the Gershenson study, so too did college persistence rates. The National Student Clearinghouse Research Center, the research arm of the organization that serves as a warehouse of educational information related to students in all grade levels, reported that slightly more students (a 2.6% increase since 2009) in North Carolina completed the first year of college and persisted into the second year. More students were graduating from high school, and more students were then staying in college between their first and second year of college. During this time, test scores were flat, which draws attention to the ability to rely on standardized assessments and college entrance exams. What are they testing if they cannot predict an increase in college success?

In my previous blog post, I pointed to the research that freshman year grade point average has served as a better predictor of future academic success than the ACT scores that Gershenson analyzed. Fordham Institute’s Michael Petrilli noted this in the executive summary of the Gershenson report, writing that “[m]uch research shows that students’ cumulative high school GPAs—which are typically an average of grades in twenty-five or more courses—are highly correlated with later academic outcomes.” GPA is not just correlated with future success; the link is much stronger than that. The University of Chicago Consortium on Chicago School Research has further shown that the GPA at the end of 9th grade, when many of these students in North Carolina took Algebra 1, is an accurate predictor of future academic success.

Leave a comment

Neither the state assessment nor the college entrance exams are the best available tools to pick out which students are likely to do well in the first year of college. Algebra 1 is a course that many students take during the 9th grade, a common time to be in the first year in a new school. We should trust the grades assigned by the teachers before jumping to conclusions about how well students are prepared for high school graduation or college based on a measure generated from a single-day event like a standardized test score.

Organizations like the College Board, makers of the SAT, produce a steady stream of reports and studies that suggest that their test scores remain flat and that this is proof that grade inflation is a pernicious problem, falsely signaling that students are ill-prepared for college. However, the test makers have a vested interest in ensuring their tests remain the prevailing measure of student success and preparation for college. Rejecting the grades given by teachers who spend more than 9 months out of the year with a student devalues the professional judgment of the teacher in favor of a company that hands out a packet of paper to the student for one morning’s work.

In 1785, then-president of Yale University, Ezra Stiles, which of the 58 seniors that year were Optimi (Latin for “best”), second Optimi, Inferiores (“lower”), and Pejores (“worse”)1. This account represents the first recorded attempt to assign grades to students. Later at Harvard and Mount Holyoke College, letter grades were given out to students before the letter grading system became as synonymous with education as the chalkboard or the teacher in the front of a classroom.

Letter grades students earn at the end of the term are a surprisingly accurate measure of student performance as well as a predictor of which students are most likely to succeed in the future2. However, final letter grades are fundamentally subjective. This is what I call the paradox of grades: they are objectively accurate even if grading itself is fundamentally subjective. What is it about grades that allows us to rely on them to predict success? More simply, what exactly are grades?

Subscribe now

Among other things, letter grades are a measure of…

How well a student mastered the content of a subject

Students receive a grade at the end of a term, with certain grades indicating mastery (an A) of the subject, while other grades (B, C) suggest the student has done a good or merely an adequate job of demonstrating mastery of the subject. There are no hard or fast rules to this grade scheme. A student with a C- might have done very well on the tests but failed to turn in homework or vice versa. Each teacher ultimately is making the call about the grade a student earns in that teacher’s classroom.

The degree to which a student met expectations

Teachers can have any number of assignments, quizzes, labs, tests, or extra credit assignments in their courses. The letter grade a student earns can be made up of some or all of these components. What the letter grade is telling us, then, is the number of hoops a student has jumped through by the time the course is complete. A student who has performed poorly on a test may be allowed to turn in extra credit to reach a higher final letter grade. What the final letter grade represents in this case is that the student has done enough to earn the grade they desire by following the steps the teacher has laid out for them. Some might refer to this as the ‘playing the game of school.’

The relationship between a student and their teachers

The term “teacher’s pet” belies a truth that we all know. Some students are the favored ones in a classroom. These students can rest easy knowing the teacher will give them the benefit of the doubt when the time comes. Students who are well-liked by their teachers earn a final letter grade that is reflective of the relationship they have with their teachers. We can further assume that students with a high grade point average are almost certainly the kind of students that most teachers enjoy having in their classes. That being the case, the individual letter grades tell us something about the individual relationship, while the sum of all grades tells us a great deal about how the student gets along with all their teachers.

Sustained effort over time

The final letter grade of a marking period is typically cumulative. Students earn a portion of their final grade at many points throughout the term by completing assignments, participating in class, and completing tests and quizzes. Doing well on just one of these components is often insufficient for the student to earn a good grade. Students have to demonstrate sustained effort, turning in homework, taking part in class discussions, and the like. While a single test will tell us how well a student was able to perform at that particular moment in time, the final letter grade is informed by many moments.

Share

For all of these reasons, we can look at grades and draw conclusions about them and the student. A student with a history of receiving mostly A’s and B’s, for example, is a student who has generally done the things their teachers have expected of them. It is also a safe assumption that the student gets along with most of their teachers. Consistent good grades can tell us which students do well in a school environment. This might explain how the grade point average during the 9th grade year can be more predictive of success in the first year of college than standardized test results such as the ACT or SAT. The students who have figured out how to navigate school during their first year in a new building are most often the ones who are going to navigate the first year of college.

A student who experiences a dip in grades, such as a sudden onset of poor marks, also tells us something. That student needs attention. Something has gone wrong in the classroom or in their social or personal life. While we cannot possibly know what it is that has gone wrong just by looking at the transcript, we do know that something is awry. In this way, letter grades can be a sort of thermometer for students, telling us about the academic record as well as the relationship between a student and their teachers. This measure is not particularly detailed, but it does signal which students need attention. In a school filled with hundreds or thousands of students at a time, with a myriad of needs, schedules, activities, and events going on at the same time, a measure that enables educators to pick out which students need more attention and support can be a valuable tool.

Leave a comment

Moreover, letter grades are routinely assigned, handed out at the end of every grading period. Unlike the annual standardized test scores that are the prevailing measure of academic success in the United States, letter grades are almost immediately available at the end of each grading period as well as the school year. The teachers are the ones giving out the grades, after all, instead of some outside company that has to analyze results for thousands of students in one batch. Individual schools and districts wait months for standardized test scores to be made available by the state department of education. This process is baked into federal regulations for reporting school performance. Those test results are delivered back to the school in the fall, an entirely different academic term than the one when the test was administered. Teachers are no longer teaching the same students when the test results come back.

A letter grade from the previous school year is available to analyze and scrutinize over the summer before students return in the fall. Teachers are placed under a strict timeline for delivering final letter grades to the main office so that report cards can be issued. Students receive letter grades in the middle of the school year as well. They are assigned at the end of the first marking period, allowing the teacher to adjust instruction and support in the middle of the year. The letter grades students receive are routinely assigned and available to schools, parents, and students to paint an academic picture of each student, for each class, and for their academic experience in total. If only we would look at them with more care and attention.

Note: Here is the new ending to the blog.

Artificial intelligence systems that grade student performance remove all of these relationship-dependent signals that we get from letter grades. As long as education takes place in social settings, we risk losing vital information by turning over grading to algorithms and machines. As I have written previously, keeping a human in the loop might be necessary even when AI expedites the busywork aspects of teaching. AI systems hold great potential when they help summarize and present data from student grades. If I had to guess, my hunch is that AI systems that flag students when grades suddenly dip will be particularly useful for schools.

Subscribe now

This led me to ponder if AI is intelligent in the way that humans reading this blog post might think of that term. If you ask the latest version of large language models (LLMs) to perform a task, they often will, and cheerfully (much like the collective hive-mind characters in Pluribus). You can ask an LLM to read and summarize your emails; prepare a work plan; find contact information for potential customers; automatically respond to social media posts; or plan a trip to the Amalfi Coast that highlights local favorite dining spots. All of this mirrors human intelligence in a way. This led me to ask ChatGPT its “thoughts” about being more intelligent than humans.

Let me spoil the lead and share the response to my question:

Short answer: no — not in the way we usually mean “more intelligent than humans.”

Longer answer: LLMs are different kinds of cognitive systems, with strengths that can exceed humans in narrow domains and weaknesses that are fundamentally unlike human intelligence.

How polite. I do think this gives us a clear division between knowing things and being intelligent (the concept of general intelligence). LLMs can quickly access and summarize information in ways that humans cannot. If you gave an LLM an IQ test, for example, it would score off the charts in certain domains. Vocabulary would pose no challenge. Tests of LLMs show that they perform poorly on novel problem-solving tasks or visuospatial reasoning. Having access to the sum of human knowledge is not enough to be generally intelligent.

Leave a comment

The Radiolab podcast explored the nature of general intelligence through a series called “G” a few years ago. This series presented important reminders that IQ tests are flawed instruments that have been used to advance repugnant theories, such as eugenics. As we grapple with the meaning of intelligence, grounding our understanding in the context of our struggle to define it is an important safety rail. We can easily access information better than at any other time in human history. Do we have the capacity to apply it in the real world? Perhaps the real test of intelligence, then, is the ability to apply knowledge to problems in the real world.

Share

Subscribe now

Take, for example, Alpha School. This private school has recently expanded to 14 sites around the United States after its initial launch in Texas. The $65,000-per-year school has embraced the use of AI to its core. There are no teachers, only guides. The school claims its 2-hour learning model results in twice the academic gains in a fraction of the time (independent research supporting this claim is difficult to find). Students receive their core academics through a proprietary AI-enabled tutoring program that individualizes the educational experience for every student. After the 2-hour academic blocks are complete, students are free to explore project-based learning activities as they interact with their guides and other students.

Walking the halls of tech-focused education conferences, you will find tons of programs that put AI right in front of the students. The major AI systems from OpenAI, Anthropic, Google, and others come with tutoring modes for students to ask questions and receive individualized support. Early evidence of generative AI use by students without the presence of significant guardrails has not been encouraging. Perhaps this model, where students are given direct and almost unrestricted access to AI, is not a good solution. There is another model that might be a better approach.

Share

Compare Alpha School’s 2-hour learning model to Paloma Learning, where AI is used to support human tutors. Paloma uses generative AI to create research-based instructional materials that teachers send to parents. The parents then deliver tutoring on literacy and math at home with their children, with support from the program. This is a so-called “human in the loop” model, where a human is part of the process and can have a hand in the output of the AI system. One early study has found that the AI + human tutoring approach can be just as effective as a human-only tutoring approach.

On its surface, being just as good but not better than a traditional human-only tutoring approach seems like a non-starter. If humans alone are just as good, why bother? The difference is that a human + AI approach can recruit thousands or even millions of people to become tutors with limited support from teachers. Paloma Learning is making a bet that being able to assign tutoring to parents and giving them AI-generated instructional materials will result in better academic gains than AI-only tutoring.

Leave a comment

Unfortunately, it is too early to know for sure if the human-in-the-loop AI models are better than the widely available direct-to-student models. Rigorous research conducted in the real world takes time. I will admit that I have placed a personal bet on the human in the loop model. Eddie, my company’s product, is designed to give teachers answers to questions about what works in the classroom.

I am assuming that educators are best positioned to make the correct decisions for their students. Eddie speeds up the process of researching what works or creating instructional materials such as lesson plans or assessments. The human in the loop is the educator who delivers the instruction. The fundamental question of the moment for AI in education is whether the technology is best when it faces the student or when it supports an adult.

For more than 60 years, the U.S. Department of Education has run the Regional Educational Laboratory (REL) program. It is written into federal law and was the subject of a federal court ruling earlier this year after the Trump Administration abruptly cancelled the contract for them and other programs. My former employer held the contract for one of these RELs. The “Labs” are anything but the image you might have of a sterile environment where new teaching techniques are tested out in controlled environments while researchers wearing lab coats take copious notes. In the same way, the “science of education” is also a misconception.

Subscribe now

The science of education is a play on the popular term, the science of reading, that has gained widespread attention in the wake of the Sold a Story podcast. The podcast takes a close look at why so many schools in the United States and abroad have adopted early literacy practices that have almost no evidence that they were effectively teaching young children how to read. In response, the science of reading is a catch-all term to describe practices that have proven to be effective, such as–but not limited to–phonics-based instruction. The issue is that science is not settled.

Science is a process that includes formulating a hypothesis, testing assumptions, collecting and analyzing data, and reporting the results in the context of the literature on the topic. I think what is meant by “the science of education” (or reading) is that there exists a set of practices that have such overwhelming evidence behind them that they are settled. There are too few practices that meet that standard. And even when they do have that weight of evidence behind them, some are ignored, as was discussed in the Sold a Story podcast.

Share

Conducting research in school settings is an incredibly complex and expensive thing to do. Students, teachers, and administrators are rarely in the same place for very long. Kids get sick on important days when surveys are handed out. Teachers and administrators quit or go on leave. Parents must be notified of their child’s participation in a study. Districts need to sign off on participation. Everything would be a lot easier if this all took place in a sterile laboratory.

Leave a comment

The popularization of the phrase, the science of education, is a reflection that children should be exposed to practices that have a heaping pile of evidence behind them. The approach is better than the alternative, relying on stubbornly persistent educational myths that may actively harm students. With Eddie, we think we can help bridge the gap between what the research says and the reality of what goes on in the classroom. We believe every educator should have access to the same insights that researchers are discovering and adding to the science. The science should not be buried in academic journals, but delivered in plain language, with clear takeaways. Science is a dynamic process that changes all the time as new studies are added to the literature. That’s the science of education, and that’s what Eddie is bringing to life.

Ben Kroll is the CEO of the Art of Problem Solving, a curriculum provider with a focus on making math fun and engaging. He used to run EdTech Insiders, so he has keen insight into how technology is being used in the classroom. I was able to ask Ben during his panel discussion Q&A about what he thinks pubic education looks like in 10 years. Ben shared that print sales are up 40% for his company. While students and at least a few teachers and parents are rushing to AI, there is pushback. That makes sense in a climate where states and districts on both sides of the political spectrum are banning smartphones in schools. Even if AI is everywhere all of a sudden, there will be those who want to get back to the core of teaching and learning. This begs the question: Does AI solve a problem in education, or is it just the latest thing?

Subscribe now

AI isn’t a buzzword for Hunters Point Research Technologies; it is a necessity. The sheer volume of educational research is staggering. Thousands of studies, briefs, guides, toolkits, and infographics are published every year. No single human can keep up, let alone translate it all into parent- and educator-friendly guidance, tools, and resources. The research is also in competition with parent and teacher blogs, influencers, social media, and many other types of content that are less likely to be grounded in evidence. Sifting through all of this content is an impossible task for any individual who needs to make a decision between department meetings or after-school activities.

Leave a comment

This is one of the core problems that AI solves for us. We use it to scan, summarize, and synthesize research at scale. Parents and educators are swamped with things to do, so helping them stay up with the research and translate it into tools and resources they can use at home and in the classroom is the main service we provide. It just so happens that AI allows us to do this in a way that has never been possible before the latest generation of large language models became available.

Share

What I also heard at Columbia University, across multiple panels and hallway discussions, was the importance of centering humans in this work. AI can do tedious work in an instant, but it cannot replace humans in school or at home. Our goal at Hunter’s Point Research Technologies is not to replace parents, educators, or researchers. We’re amplifying their work, making it more accessible, more actionable, and more impactful for the families and educators who need it most. We are making a bet that parents and educators want to use AI to give them the time they need so that they can focus on helping young minds grow and develop. The point of AI, then, is to allow all of us to focus on what matters.

In co-creating Eddie (available to the public beginning on November 3rd - join the waitlist today), my business partner and I are making a bet that parents and educators experience problems with the status quo in finding and using evidence at home and in the classroom. Our solution is based on many assumptions, but that is the core problem we are trying to solve. On the surface, it should not be a problem.

Subscribe now

Much of the research on what works to improve academic outcomes is freely accessible right now. The U.S. Department of Education’s Institute of Education Sciences–the research arm of the federal agency–has two fantastic resources that make research available (as long as the current administration maintains them): the Education Resource Information Center (ERIC) and the What Works Clearinghouse (WWC). Both serve slightly different purposes but offer free access to thousands of peer-reviewed journal articles that form the basis of the literature on improving academic outcomes.

In my previous job, I interviewed educators about the WWC. We found that most teachers were unaware of the WWC and its various products. Those who did know rarely used them. These resources are targeted to educators, so parents are even less aware of them. Many parents whom I have interviewed lately use Google or other non-academic sources, such as friends, parenting podcasts, TikTok, or other social media. There are several reasons for this disconnect.

Leave a comment

By its very nature, academic articles are not written for practitioners. They include sections most laypeople would find unhelpful, such as a description of methodology that goes into great detail on the statistical approach used to conduct the study or conclusions that discuss the limitations of the research. If they describe the academic program, policy, or practice that was studied, details on how to implement are sparse. Worse still, these papers are filled with jargon intended for fellow researchers. One recent example came from my own household, where I searched for articles about ways to get my toddler to stop having tantrums. I discovered the academic community referred to these as “extinguishing strategies,” as if that was an obvious search term.

Even in the rare cases where practitioners are aware of the research, they lack the time to read through articles one by one. Teachers are estimated to work an average of 54 hours per week during the school year. Parents have to juggle home life and often two careers. The people most in need of the research are short on time and more often than not lack the specialized knowledge and skills to interpret what they find.

Share

As I have conducted interviews with dozens of parents and educators over the past 6 months, they routinely tell me that they lack the time to find the research, and even when they do seek out the best approaches, they do not trust their current options. Evidence clearinghouses such as ERIC or Google Scholar take too much time to read and require training to interpret what they find there. Teachers are adopting large language models such as ChatGPT and Claude to perform administrative tasks at staggering speeds. Many of the educators who talked to me said they distrust what they receive. A better tool is needed, one that speeds up the process of sifting through the research while translating it into terms that everyday people can understand and use.

For Eddie, we are assuming parents and teachers want to use evidence to make better decisions for their children or students. One of the other major assumptions underlying our work is that we can solve this problem, to build a system that instantly reads the hundreds of thousands of peer-reviewed journal articles to answer questions in plain language and then turns those answers into ready-to-use resources that parents and educators can use at home and in the classroom. As a researcher, I am excited to soon find out if my assumptions are correct.

For the past 15 years, I have been an applied researcher in education and social sciences. Throughout my time leading and working on research projects, there was a problem that kept showing up in conversations with parents, educators, and researchers: finding trustworthy, evidence-based practices for K-12 education was harder than it should be.

As someone who has worked across education and research, I saw the disconnect firsthand. Parents were in the dark about what happened at school. Teachers were overwhelmed with teaching classes, preparing lessons, attending meetings, grading assignments, managing student behaviors, and more. Administrators were under constant pressure. And researchers, the ones generating real insights, were writing and publishing their findings in ways that everyday people couldn’t use.

Leave a comment

If the general public were aware that a new study had been published - most weren’t -, it would be written in a manner that buried the lead. Hidden behind sections on methodology or conclusions that explored areas for further research rather than practical advice, the real insights were about how and where the program, practice, or policy worked. That valuable information was hardly discussed in the research report.

These are problems that have existed for a long time. Professor Thomas Dee of Stanford University wrote about this for EdWeek earlier this fall. Mark Schnieder, former director of the Institute of Education Sciences (IES) at the U.S. Department of Education, wrote about this in 2018. The What Works Clearinghouse was founded by IES to address this very issue more than 20 years ago. Yet the problem persists.

So I founded a company to solve this problem. Not as a tech experiment, but as a bridge. A way to connect families and educators with vetted, practical strategies grounded in science. Starting a company wasn’t the goal; Empowering people with better tools was the goal. The company, Hunters Point Research Technologies, Inc. (HPR Tech), is my way of making that happen.

Share

There are many reasons that I should not have started HPR Tech. This is a huge risk. I have young children (3.5 years old and 6 months as I write this) and a family to support. My partner fully supports me, but we both know the downside risks. How long can we afford to go without an income - in New York City, no less - with rent, daycare, startup, and other expenses piling up?

Earlier this year, thousands of highly-qualified professionals in education and other industries were let go through mass layoffs in the federal government and reductions in force from federal contractors. I was included in one of those rounds of layoffs in March. The job market was awful. Every job, even for highly technical positions for experienced staff, was being met with hundreds, if not thousands, of overly qualified candidates. Business leaders were even complaining (or commiserating?) about the situation on LinkedIn.

As I slowly worked my way through job applications all day and waited for emails that came months later, regretfully notifying me that the employer had gone in a different direction, there was a loud voice in my head that told me that what I was doing was taking me far from my passion. Some of the companies were doing interesting, even inspiring things. As time went on, my standards shifted, and I felt further disconnected from the job descriptions. It wasn’t for me.

Subscribe now

I am not going to lie and say that the idea of being in charge doesn’t motivate me a little. Still, the main thing that gets me going is solving a problem. There is a drive inside of me to turn my vision into reality, to solve problems. The problem I have chosen is the same one that I addressed when I worked for a large non-profit research organization. I am dedicated to getting evidence into the hands of the people who most need it. My career is focused on making better decisions by following the evidence.

Emerging technologies, including AI and machine learning, are creating opportunities that previously never existed. We can digest and share information at speeds that were never possible before. I am fortunate that I have friends and colleagues who are much smarter than I am who can help me figure out how to pull this off. The path is different than what I had planned, but the mission remains the same.

While I do not have the technical or business background to run a company, here I am. I’m excited to share this journey with you. I would love to read your thoughts about what drives you.