A few weeks ago, we released a new white paper, Beyond Literacy and Fluency: Preparing for AI Agency and Stewardship. The paper presents some of our latest thinking about the progression of capabilities, organizational practices, and leadership attention necessary to make AI a sustainable part of an organization’s work.

Here’s a little background on how we got there.

We follow research across several fields to inform our work with organizations implementing AI. Interestingly, we have found that research on organizational development, implementation, and mission alignment often provides more useful direction than the still relatively limited research on sustained, scaled AI integration.

That isn’t to dismiss the growing body of AI research. But much of what is being published focuses on individual tools, short-term outcomes, or relatively small implementations. There is considerably less evidence about what it takes for organizations to integrate AI responsibly, sustain that work over time, and demonstrate meaningful improvements.

Through our work with organizations developing policies, guidance, implementation plans, and innovation priorities, we have learned quite a bit about what it actually takes.

And it takes much more than AI literacy.

Here is an example from our organizational work.

About 18 months ago, we began working with a nonprofit organization that was eager to develop its team’s AI skills. The initial interest was largely focused on training, helping staff understand available tools and use them more effectively.

Instead of beginning with training, we started by examining what was already happening within the organization.

Through data collection, conversations, and collaborative design, we learned that almost all staff were already using personal AI tools. Nearly everyone was sharing some form of internal intellectual property through those tools, and 100% of respondents wanted clearer guidance about what they should and should not be doing.

None of this was particularly surprising. What mattered was what we did with that information.

We worked with the organization to develop guidance grounded in its actual work, then stress-tested that guidance against its innovation priorities and intended use cases. We developed a training plan to strengthen AI literacy, established a leadership agenda to support continued AI fluency and purposeful decision-making, and created a three-year organizational plan focused on AI stewardship.

Each part of the work informed the others. As staff developed new capabilities, questions became more sophisticated. As use cases became clearer, policies and guidance needed to be revisited. As leaders developed a better understanding of what AI could support, they also had to reconsider priorities, responsibilities, and organizational expectations.

The process is continuous, not a series of training sessions that eventually ends.

And while this may sound technically complicated, it does not necessarily require an organization to hire a full-time AI specialist or develop substantial internal technical expertise. It does require commitment, leadership attention, and a willingness to involve the people doing the work. We use collaborative design teams to distribute responsibility, test ideas, and help organizations make progress without placing everything on one individual.

What we have seen is that AI literacy becomes more sophisticated, AI fluency continues to develop, and people become increasingly capable of questioning how, when, and why AI should be used.

Perhaps most importantly, leadership decisions are continually examined as organizational knowledge and experience grow.

That is where agency and stewardship become particularly important.

Our white paper attempts to bring together what we have learned from this work, our implementation research, and the broader research base. I hope you will give it a read. There are some practical lessons embedded in those pages.

Now, let’s talk about educator preparation

One area our team has been working to support over the past year is educator preparation. Through research, collaborative projects, professional learning, and conversations with faculty and institutional leaders, we have been paying close attention to how educator preparation programs are using AI in their own work and preparing future educators to use it.

What we continue to find is that many programs are operating within university-level policies that were not necessarily designed with educator preparation in mind. Others are waiting for K–12 systems to establish clearer expectations before making significant changes themselves.

This creates an interesting problem. Educator preparation programs are responsible for preparing people to enter a profession that is already being affected by AI, yet they may have limited authority to determine how AI is integrated into their own programs.

This week, I had the opportunity to spend time with colleagues at the Teacher Education Division (TED) of the Council for Exceptional Children conference in Denver. I sat down with faculty and program leaders to discuss what they are experiencing, where they are making progress, and what continues to get in the way.

Those conversations, combined with our work with educator preparation programs over the past several years, surfaced a few key issues I think deserve more attention.

1. Governance before tools is still largely missing from the conversation. We continue to see programs moving into tool adoption and training before establishing clear guidance about acceptable use, data protection, intellectual property, or the educational purposes those tools are intended to support. In some cases, tools are being introduced to teacher candidates without agreed-upon use cases or sufficient clarity about what information can be shared. During one conversation, a colleague described a reported multimillion-dollar copyright dispute involving curriculum materials uploaded to a personal ChatGPT account for undergraduate coursework. I have not independently verified the legal details, but the concern it illustrates is worth examining. Who is responsible for ensuring that faculty and candidates understand the implications of uploading institutional materials, student information, or copyrighted content into AI systems? And what happens when institutions encourage experimentation without first addressing those questions? Governance should not be something we develop after a problem occurs. It should help establish the conditions under which meaningful and responsible innovation can take place.

2. We need to develop mindsets, not just AI literacy. Organizations continue to focus heavily on tools and technical skills. But tools change, sometimes dramatically, while curiosity, critical thinking, professional judgment, and the ability to question assumptions remain important across technologies. Our internal data across organizational engagements (n = 284) indicates that approximately 85% of respondents do not routinely scrutinize or review AI-generated outputs. That finding concerns me. We can teach people how to write prompts, generate lesson plans, and create instructional materials. But if they are not questioning the accuracy, appropriateness, assumptions, or implications of what AI produces, we have not adequately prepared them to use it. This is not simply a knowledge problem. It is also a matter of habits, expectations, and professional judgment. And those do not develop through a single workshop. We need to create opportunities for educators and candidates to practice evaluating AI outputs, challenge recommendations, identify when something is inappropriate, and make intentional decisions about whether AI should be used at all.

3. We still have not established a shared understanding of quality AI integration. Across our higher education engagements over the past 18 months, we have yet to encounter a team that has established a comprehensive, shared definition of what quality AI integration looks like within a course, program, or educator workflow. Some programs have developed shared language. Others have identified tools, created resource collections, or established general expectations for responsible use. Those are useful starting points, but they are not the same as defining quality. What does high-quality AI integration look like in a teacher preparation course? How would we recognize it in a clinical experience? What evidence would demonstrate that AI is improving candidate preparation rather than simply making coursework more efficient? And how should those expectations differ depending on the learning outcome, the candidate’s existing knowledge, or the professional responsibilities they are preparing to assume? Much of the available guidance continues to focus on tools or discrete skills. Programs need something more substantial: a shared understanding of what effective, purposeful, and responsible AI integration actually looks like in practice. Without that clarity, faculty are left to make individual decisions, often without the expertise, time, or institutional support necessary to evaluate those decisions.

4. There is still a gap between faculty’s personal use of AI and their use of AI in teaching. Faculty are using AI for research, course preparation, writing, planning, and other professional tasks. But those experiences are not consistently translating into opportunities for teacher candidates to develop their own capabilities. In our conversations, faculty frequently described feeling stuck. Some are using AI extensively in their personal work but remain uncertain about whether they should introduce it into coursework. Others are holding back because institutional policies are unclear, disciplinary expectations are still developing, or they are unsure what constitutes good practice. This creates a situation in which faculty may be developing considerable personal fluency while candidates receive relatively little structured preparation. And again, we come back to organizational clarity. Faculty need more than permission to experiment. They need shared expectations, opportunities to collaborate, examples of quality integration, and support for examining how AI changes the work they are preparing candidates to do.

5. We need to reconsider how much responsibility we place on a single champion. In previous educational technology initiatives, it was common for one enthusiastic faculty member or a small group to carry much of the work. We are seeing similar patterns with AI, but I am increasingly concerned about whether that approach is sustainable. In some programs, champions have taken on so much ownership that others have difficulty participating or developing their own capacity. In other cases, enthusiastic faculty have the knowledge and commitment to move the work forward but lack the organizational authority to make meaningful changes. Neither situation is particularly effective. AI implementation affects curriculum, assessment, clinical preparation, institutional policies, professional learning, and relationships with K–12 partners. That work cannot reasonably sit with one person. Several faculty members also described the consequences of institutions taking too long to make decisions. They recognize that programs need to move, but competing priorities, uncertainty, and fear of making the wrong decision can leave institutions almost paralyzed. We need leadership structures that distribute responsibility, create meaningful opportunities for participation, and give faculty the authority and support necessary to make progress. Champions can help initiate change. They should not be expected to sustain it alone.

6. We need to move beyond early adopters while making room for non-adoption and resistance. Much of the professional learning being offered in educator preparation remains informal or optional. Not surprisingly, those opportunities tend to attract faculty who are already interested in AI, relatively confident using it, or actively experimenting with available tools. That leaves a significant portion of faculty outside the conversation. Some colleagues told us they were reluctant to spend time exploring AI without clearer institutional commitments or an understanding of what the work was intended to accomplish. Others raised serious ethical, professional, or personal concerns about AI use. I think those concerns deserve meaningful attention rather than being treated as barriers to overcome. There is a difference between someone who lacks confidence using AI, someone who does not see its relevance to their work, and someone who has principled objections to particular uses. Those positions require different conversations. At the same time, some institutional leaders are increasingly focused on getting faculty to use AI, sometimes without articulating why, for what purposes, or what quality implementation should look like. Simply telling faculty to use AI is not an implementation strategy. Some leaders are beginning to connect AI integration to accreditation, program improvement, or other institutional priorities. That may create opportunities for more coordinated action, but those efforts are still relatively limited. We need to make room for critical examination, including decisions not to use AI when its use does not serve a meaningful educational purpose. Agency includes the ability to make those decisions.

7. Teacher candidates do not have unlimited time to wait for institutions to figure this out. This may be the issue that concerns me most. Teacher candidates have a finite period of preparation before they enter classrooms and assume responsibility for students. Meanwhile, AI is already influencing work across business, government, healthcare, education, and other sectors. Organizations are developing new practices, documenting efficiencies, identifying limitations, and reconsidering how people and technology work together. Educator preparation programs are facing growing pressure to respond, and that pressure is coming from multiple directions. Faculty are trying to understand what candidates need to know. K–12 partners are making their own decisions about AI. Institutional leaders are managing competing priorities. And candidates are entering a profession in which expectations for AI use are still being negotiated. Changing systems takes time. Changing professional habits, instructional practices, organizational expectations, and decision-making processes takes even more deliberate effort. It also requires resources. We cannot reasonably expect faculty to redesign coursework, reconsider assessment, develop new expertise, collaborate with clinical partners, and establish responsible practices without time, support, and organizational commitment. But waiting indefinitely for perfect clarity is not a reasonable option either. Educator preparation programs have an important opportunity to help define what responsible, high-quality AI use should look like in schools. They should not have to wait for every K–12 system to make those decisions first.

A few final thoughts

I left this week thinking about how much educator preparation programs have in common with the other organizations we support.

The tools may be different, the responsibilities may be different, and the institutional structures are certainly different. But many of the underlying challenges are remarkably similar.

People need clarity about purpose. They need opportunities to develop knowledge and skills. They need guidance that reflects the realities of their work. They need meaningful participation in decisions about implementation. And they need leadership that understands this is an ongoing organizational effort, not a temporary technology initiative.

That is part of why we developed the progression from AI literacy and fluency to agency and stewardship.

Literacy helps us understand AI. Fluency makes us capable of using it. Agency makes our use more purposeful. Stewardship helps make that use trustworthy and sustainable.

These capabilities are interconnected, and they continue to develop as people and organizations learn.

For educator preparation, I think the implications are particularly important. We are not simply preparing future educators to use the tools available today. We are preparing them to make decisions about technologies that will continue to change throughout their careers.

That requires more than technical proficiency. It requires the ability to question, evaluate, adapt, and make decisions grounded in professional knowledge, educational purpose, and the needs of learners. And it requires institutions willing to do the same.

I hope these conversations continue, and I would be particularly interested in hearing how other educator preparation programs are approaching these issues. What is working? Where are you getting stuck? And what have you learned from implementation that others might benefit from knowing?

There is a lot we can learn from one another, including the things that have not gone as planned.

Interested in having Jody present at your event or spend time with your team? Learn more and contact her team at JodyBritten.com. Remember that the future of education is being shaped now. We need informed, thoughtful voices to make meaningful choices. Subscribe to Jody’s newsletter for weekly updates as soon as they’re published at https://jodybritten.medium.com/subscribe

Globally, it was no secret that the midterm elections in the United States could have international consequences. Over the past two years, we have seen… well, a lot of things.

On a basic human level, much of what we have witnessed in American politics has been less than exemplary. The misuse of power, pay-to-play politics, angry and hateful rhetoric, and what I can only describe as obnoxious behavior focused on attacking one side or another have become far too normal.

At the end of October, the U.S. Department of Education shared a series of mock Halloween costumes on its official social media accounts. The images made caricatures of educators, inclusion, and people associated with various educational and political perspectives.

To say I considered this a low point is an understatement. But undercurrent was all too familiar: We are right. You are wrong and we will spend public time and funding to prove the point.

And even then, I found myself wondering just what we were spending our time and energy on — because arguing one’s point seems to be the default that doesn’t capture much momentum. Out of curiosity, I used a few handy AI marketing, and language analysis tools this week to look at the tone, language, and sentiment associated with candidates who won and lost their elections.

This experiment was 100% an informal analysis, and by no means even coming close to being a comprehensive study of voter behavior. But something stood out. The messaging that the data suggested was most compelling was not focused on party loyalty or political control. It was about public goods, public systems, trust, accountability, and the checks and balances that are supposed to keep our government functioning. And I think that matters.

The election results were certainly a significant victory for Democrats. But I don’t read this election as an overwhelming endorsement of Democrats as a party. I see it as an overwhelming rejection of the direction of the current Republican leadership and a demand for greater accountability and balance.

Let me be clear. I firmly believe that any political party left unchecked can fundamentally undermine our democracy. That is precisely why checks and balances matter.

As I spoke with family and friends over the past several weeks, I kept coming back to how exhausted I was by the political advertising. So much of it was organized around fearing others, blaming specific groups of people, and convincing us that someone else was responsible for everything going wrong.

I kept waiting for an advertisement in which a candidate simply said, “Here is my plan.”

Not just what they wanted. Not who they disagreed with. Not just what they would fight for. Not another reason we should fear the other side. But how were they going to make things better? That sadly seemed largely missing from my news feeds.

And if you know me at all, you know I am a firm believer that how is the hard part. Anyone can tell you what they want to accomplish. Implementation is where things get real.

So now we find ourselves at a moment when new leadership will arrive in communities, states, and Congress in January. My hope is that we keep our eye on the ball.

We have had leaders so committed to being right that they have lost sight of the human consequences of their decisions if they are wrong (or flat out didn’t care what humans were voided, harmed, made to suffer, etc.). And I worry that some incoming leaders may arrive guns blazing, focused more on winning the next point in the debate than delivering meaningful change.

That would be an enormous, missed opportunity.

We have a chance to reset how we work together, how we nurture community, and how we approach being, first and foremost, good humans that together share a country, land, and resources that ourselves, our kids, and our families depend on.

And that brings me to public education.

Maybe this is our moment to rethink public education

I hope we take this moment to reaffirm our national beliefs in the importance of public education.

I have worked with educators on six continents and in all 50 states at this point in my career. I have seen firsthand the competing priorities, financial challenges, missed opportunities, extraordinary successes, and yes, the considerable waste of time and energy that can exist in our education systems (for both adults and kids).

I have seen what works. I have seen what doesn’t. And I have seen what happens when talented educators are given the resources, trust, and opportunity to do things differently.

It has been hard to watch funding bleed away from the schools my own kids attend. It has been hard to watch teachers who want to innovate get pulled into systems so consumed with test scores that the measures themselves become more important than the learning they are supposed to represent. It has been hard to see decreased attention on creating community within schools in favor of iron rule and systems of control.

And it has been exhausting to watch people suggest that teachers are somehow spending their days indoctrinating children, promoting gender transitions, or introducing evil ideas like loving your neighbor through picture books that have been sitting in our homes, schools, and libraries for generations.

Seriously. Go teach kindergarten for a day and then tell me how much time you think educators have for these supposed conspiracies.

Fear has led us deeply astray in how we think about public education.

But I also think we need to acknowledge something that may initially seem uncomfortable: While the political battles have consumed enormous amounts of attention, public education itself has too often remained stuck in policy cycles that prioritize accountability through testing rather than asking whether the learning experiences we provide remain relevant to the students sitting in our classrooms.

We have spent decades debating how to improve scores while too often failing to reconsider what school could actually look like for the kids that we serve.

I wrote about this a few weeks ago in We Built the Wrong Version of EdTech. Today Our Students Know It. The fundamental argument remains the same.

Our students have changed. The world has changed. The opportunities for learning have changed. But much of the way we organize, and measure education has not.

The entire point of public education is to ensure that the population is educated. That sounds simple but think about what it actually means. It does not mean that every child must learn in the same place, at the same time, in the same way, using the same resources. Nor does it mean every student should have an identical educational experience, nor a limited educational experience based on their zip code.

We have hundreds of ways to organize meaningful learning, many of which were not even possible ten years ago. And yet we continue to operate much of public education as though the structure itself is the thing we are trying to preserve.

What if we changed that?

What if we brought meaningful funding back to community schools, gave educators both the opportunity and responsibility to innovate from within, and began measuring what matters most to learners and their communities?

What if we reaffirmed that public funding carries public responsibilities, including accessibility, inclusion, transparency, and accountability for every organization? What if we finally took a serious look at how we use time, physical space, digital environments, community expertise, and the extraordinary resources now available to support learning?

Maybe we could ignite new models, new energy, new partnerships, and renewed public support for education. But I don’t think funding alone gets us there.

The shift away from supporting public education has not come solely from policymakers. It has also come from people who have lost trust in the system, questioned its relevance, or simply stopped believing their community schools can deliver what their children need.

That trust will not return because we tell people public education is important or has a legacy that they should believe in. My friend Emily recently posted about trust and how important it is to consider (her words mean more today, given this election).

But many researchers believe that trust will return when people in all communities can see, participate in, and benefit from public education that genuinely works for their individual children.

Almost 30 years ago, my dad and I were driving somewhere and passed a new school under construction. I remember him looking at it and saying something along the lines of, “I’m not sure school will be a singular, physical place for too much longer.” Clearly, his timeline was off. But interestingly enough, this same sentiment continues to come up in national education conversations nearly every week — among people who love public education and fight for the opportunity and system tirelessly every day.

But we have to realize that while school as we have known it might be what we want for our kids, it might not be what our kids actually need. And maybe, just maybe, we can use this moment to make some of what has long been possible a practical reality for our communities.

That starts with funding, trust, and measuring what matters most. It also requires educators, policymakers, families, and communities to be willing to reconsider some long-held assumptions.

Our kids aren’t us.

And they certainly aren’t robots who should simply show up, comply with instruction, complete their assignments, and leave. Few adults would thrive in that situation today — neither will our kids.

Learning science matters enormously. Evidence should inform our decisions. But research should help us design richer, more effective learning experiences, not become an excuse to ignore the complexity of students, their communities, and the world they are preparing to enter.

We should be asking what our young people can do with what they know, how they develop judgment and agency, and whether the experiences we create prepare them to contribute meaningfully beyond school.

This election’s results might give us the opportunity to ask better questions about public education.

Game on, people. It’s time to show up strong.

And then there is AI

Of course, this election will have implications for AI, not just in education but across society. And I think this is another place where the message from communities deserves our attention.

Concerns about data centers, infrastructure costs, resource consumption, protections for consumers, and governance were increasingly visible throughout the campaign.

In our area, I started seeing advertisements promoting data center development about three months ago. As the election approached, those advertisements became more frequent. Much of the messaging centered on what AI infrastructure investment could do for a community. Jobs. Economic development. New opportunities. Those are legitimate considerations.

But those messages were increasingly challenged by concerns about electricity prices, public infrastructure, environmental impacts, and who would ultimately pay for the enormous resources required to support this technology. Some advertisements even featured customer recordings of conversations with power companies questioning rising consumer costs.

And this was not simply a local-level debate. Across the country, voters and candidates were raising questions about data center development and its effects on communities.

To be clear, not every increase in electricity costs can be attributed to AI. But communities are right to ask whether they will carry costs created by infrastructure they had little opportunity to influence.

The physical infrastructure required to support AI is not something we can ignore.

We have been so focused on what AI can do, what it might become, and how quickly we can scale it that we have not adequately established the public commitments necessary to ensure its infrastructure is developed responsibly.

And for me, that is a fundamental governance issue.

• Who decides where data centers are built?
• Who benefits financially?
• Who pays when additional energy and water resources are required?
• Who evaluates the environmental consequences?
• What accountability exists when public infrastructure supports private technology development?
• And perhaps most importantly, how much actual influence do communities have over decisions that could affect them for decades?

These are not anti-AI questions. They are the kinds of questions responsible innovation requires us to ask.

There is a significant difference between being supportive of AI development and believing that AI companies should be allowed to scale without meaningful public oversight.

I am strongly supportive of the possibilities AI presents. My work in this field reflects that. But supporting innovation does not mean abandoning our responsibility to the public.

If anything, the extraordinary capabilities of AI make governance, accountability, and public participation more important. I would argue that this election gave us an important reminder that people want more than promises about what technology might deliver.

They want a plan. They want to know how the benefits will be shared, how the risks will be managed, and whether the people making decisions can be held accountable. And they should.

A technology this powerful needs oversight, a clear implementation plan, and meaningful public accountability. It really is that simple.

A little hope, and a lot of work ahead

It’s been a week, friends. And while I know there is considerable excitement among Democrats following the election, I hope we resist the temptation to treat this as simply another victory for one political party over another.

We need more than that.

I have friends and collaborators who won elections this week for all parts of the country, and my message to each of them has been essentially the same.

Don’t waste time beating your party drum.

Get after it.

Make change.

Build trust.

Invest in public systems.

Listen to people who disagree with you.

Work with your communities.

And focus your energy on solving real problems for real people.

The opportunity in front of us is not simply to reverse the direction of the past few years. It is to demonstrate that we can govern differently and move forward with outcomes the people want.

That we can invest in public education while also having the courage to rethink how it works. That we can support extraordinary technological innovation while protecting the communities and public resources that make it possible. And that we can disagree about politics without losing sight of our shared responsibility to one another.

I don’t expect any of that to be easy. The what is always easier than the how.

But maybe this election can be more than a political correction. Maybe it can be an opportunity to rebuild some of the trust, public purpose, and collective responsibility we have been missing.

A little hope is a good place to start.

Now comes the hard part. Let’s do something meaningful with it.

Interested in having Jody present at your event or spend time with your team? Learn more and contact her team at JodyBritten.com. Remember that the future of education is being shaped now. We need informed, thoughtful voices to make meaningful choices. Subscribe to Jody’s newsletter for weekly updates as soon as they’re published at https://jodybritten.medium.com/subscribe

A few things on my mind this week

There is no shortage of enthusiasm for AI in education. New capabilities, research findings, policy conversations, and implementation initiatives continue to arrive faster than most organizations or individuals can reasonably evaluate them.

This week, I found myself coming back to three things: evidence, enthusiasm, and agency.

We are beginning to see more continuous efforts to understand whether AI improves learning, but this technology is also challenging us to reconsider what learning actually means and how we measure it. At the same time, some of the research we have relied on so far in this conversation on AI and education may not be nearly as strong as current headlines suggest.

And there is another issue I think deserves much more attention: implementation.

I am seeing a lot of organizations talk about big wins from relatively small pilots. But anyone who has worked through an organization-wide implementation knows that the process is considerably more complicated than a few positive outcome measures might suggest. We need to understand not only whether something worked, but what it took to make it work, what did not work, what people had to change, and whether the results can reasonably be replicated.

We need to be cognizant of the fact that our traditional research and publication processes are not necessarily keeping pace with the technology, or the decisions institutions are making about it.

Meanwhile, AI systems are becoming increasingly capable of recommending, creating, and acting on our behalf. Which brings us back to agency. What should students, educators, and institutions continue to do for themselves? How do we develop the judgment necessary to make those decisions? And are we giving people meaningful authority over how these technologies are implemented?

Through all the questions, there were some interesting developments this week. A few are promising. Others should make us uncomfortable. All deserve a closer look.

What if some of our strongest evidence for AI in education isn’t so strong after all?

On October 26, The Hechinger Report published an examination of research raising serious questions about the evidence supporting AI’s educational effectiveness. One researcher conducted a methodological audit of 14 meta-analyses drawn from a larger group of 39 reviews reporting positive educational effects.

The audit identified problems with statistical methods, the studies selected for inclusion, and whether the outcomes being combined measured learning. The audit itself was released earlier and does not establish that all 39 reviews are invalid. But the reporting raises important questions about how confidently we should interpret some of the field’s most frequently cited findings.

Think about what that means for education leaders and researchers alike. We are making purchasing decisions, redesigning instructional practices, investing in professional development, and developing policies based, at least in part, on an evidence base that may be less reliable than we assume.

This does not mean AI cannot improve learning. It means we need to become much more sophisticated about what counts as evidence that it does.

There is also an elephant in the room. We have work to do to keep pace with AI in the research community. So much is happening outside traditional academic research that could provide useful early evidence, including implementation findings, rapid-cycle evaluations, and lessons from unsuccessful pilots. Yet much of that information is not systematically documented or shared. By the time some findings reach publication, institutions may have already repeated the same mistakes.

We need rigorous evidence, but we also need ways to make that evidence useful while decisions are being made.

OpenAI’s GPT-6.1 Terra: When AI moves from recommending work to doing it

OpenAI introduced GPT-6.1 Terra on October 26, extending its work on computer use, visual reasoning, and agentic workflows. The model is being introduced through ChatGPT Work, Codex, and the API, with availability depending on the product and account.

What interests me is not simply that another model has been released. It is the continued movement toward AI systems that can navigate software, coordinate tasks, and complete increasingly complex work.

There are some meaningful opportunities here for education, particularly in research administration, organizational operations, data management, and processes that require considerable staff time. I suspect much of AI’s immediate institutional value may have little to do with direct instruction and much more to do with the systems that support it.

Yet again, implementation becomes considerably more complicated when an AI system can act rather than simply advise.

Who authorizes those actions? What information can the system access? What happens when it makes a mistake? And who is responsible for recognizing that something has gone wrong?

These are not questions that can be resolved by a short training session or a policy that tells employees to use AI responsibly. Institutions will need to understand their workflows, data practices, permissions, and decision-making processes before delegating meaningful work to AI.

Google Gemini is making interactive learning experiences more accessible

Google introduced immersive explorations in Gemini this week, providing students with interactive visual experiences that allow them to investigate topics through different scales, representations, and perspectives.

Examples include exploring the Silk Road, investigating the inside of a jet engine, and examining the physics of lightning. The experiences are being introduced through Gemini’s student hub, with plans to expand how learners access them.

While digital exploration is not new by any means I do find this direction interesting because it moves us beyond the familiar chatbot experience. Students are not simply asking questions and receiving explanations. They can interact with representations of complex ideas and explore relationships that might otherwise be difficult to visualize.

But I also think we need to be careful about assuming that more engaging experiences automatically produce better learning.

What are students expected to notice, question, or understand? How does the experience connect to prior knowledge? What role does the educator play in helping students make meaning from what they encounter? And perhaps most importantly, are our existing assessments capable of capturing the kinds of understanding these experiences might support?

We should be exploring these opportunities while being equally intentional about the instructional design surrounding them.

OpenAI’s changes to ChatGPT Edu are a reminder that implementation does not end with adoption

OpenAI’s update to ChatGPT Enterprise and Edu changed how users access different reasoning levels. Automatic switching from Instant to higher reasoning levels was retired, requiring users to select an available thinking level when they need more intensive reasoning, subject to institutional settings.

This may seem like a relatively small product change, but I think it illustrates something that gets overlooked in institutional AI planning.

Organizations often approach AI adoption as a process of selecting a tool, establishing policies, providing training, and making the technology available. But the tools themselves continue to change, sometimes in ways that affect performance, user experience, administrative oversight, and the assumptions behind existing workflows.

And those changes may not be obvious to faculty, staff, or students.

This is why I continue to advocate for ongoing AI governance and implementation support rather than treating adoption as a one-time initiative. Institutions need people who understand the technology, monitor changes, evaluate their implications, communicate with users, and adjust when necessary.

The work does not end when the licenses are purchased. In many ways, that is when it begins.

Google and Stanford are helping students understand AI beyond the chatbot

Google Research and the Stanford Accelerator for Learning introduced Polar Peaks this week, adding a new experience to their AI Quests series. Designed for students ages 11 to 14, AI Quests uses a neuroscience challenge to help learners investigate how AI works and how it can be applied responsibly. Google reports that more than 400,000 quests have been completed across the series, which is available in multiple languages. Additional teacher resources accompany the new release.

I like the active learning approach in this because it positions students as investigators and problem-solvers rather than simply users of AI products.

While we have spent considerable time teaching students how to prompt AI systems, evaluate responses, and follow appropriate-use guidelines. Those are useful skills, but I don’t think they are sufficient. Students also need opportunities to understand how these systems are developed, why they produce outcomes, what they cannot reliably do, and how people make decisions about their use.

Judgement develops over time and we cannot expect students to exercise sophisticated critical judgment about AI simply because we have explained its limitations.

Students need opportunities to practice, make mistakes, ask questions, and develop the underlying knowledge necessary to recognize when something is wrong.

That is the kind of AI literacy I would like to see more schools prioritize.

Google’s classroom studies offer promise, but the implementation details matter

Google released findings from two education studies, including a ten-week randomized trial involving more than 1,100 students ages 12 to 15 in the United Kingdom. The UK study examined mathematics tutoring through Eedi, comparing static hints, human tutoring, and Gemini-supported tutoring. Students receiving AI-supported tutoring performed comparably to human-tutored students on platform assignments and intermediate quizzes.

That sounds promising, and it is. But the improvements did not translate into significant gains on an independent standardized assessment. Of course, that raises another question: Is a standardized assessment always the right measure of what we are trying to accomplish?

I am not suggesting we abandon established measures of learning, however if AI is changing how students investigate problems, receive feedback, and develop understanding, we need to consider whether our evaluation approaches capture those changes.

A separate yearlong study involving educators across 31 Italian schools examined changes in lesson design, collaboration, and teacher experience when Gemini supported planning.

Both studies are worth reading because they move us closer to understanding AI in actual educational settings. But they also remind us that positive outcomes are only part of the story.

We need to understand the implementation conditions that produced them and whether those conditions can be replicated.

OpenAI’s Model Spec puts human agency into the conversation

On October 29, OpenAI updated its Model Spec, the public document describing how its models are intended to behave. The revised guidance identifies five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. It also expands guidance related to personalization, oversight, and how models should approach decisions that may have consequences beyond the immediate interaction.

I find the explicit attention to human agency encouraging, but it also raises some questions.

Technology providers make decisions about how models respond, what they prioritize, when they intervene, and how much latitude users have to direct their behavior.

Those decisions influence the experiences of educators, students, researchers, and institutional leaders, whether they recognize it or not. So, when a provider says it values human agency, I want to know what that means in practice.

Can users understand why a system made a recommendation? Can they challenge its assumptions? Can institutions meaningfully influence how the system behaves? And do people retain the knowledge and authority needed to make independent decisions?

Articulating a value is important. Demonstrating how that value is reflected in actual use is another matter.

A final thought: Enthusiasm is not evidence, and adoption is not agency

I don’t think we need less enthusiasm about what AI could make possible in education. There are genuinely interesting opportunities here, including better instructional feedback, new forms of accessibility, more responsive learning experiences, and ways for educators and students to create things that were previously difficult or impossible.

But enthusiasm needs to be accompanied by intellectual discipline and a much stronger understanding of implementation.

We need to distinguish what AI can do from what it helps people learn. We need to distinguish positive research findings from strong evidence. We need to distinguish successful pilots from sustainable organizational practices. And we need to distinguish human oversight from meaningful human judgment.

I also think we need to pay closer attention to the capabilities people are developing as they use AI.

Are students becoming better thinkers, or simply producing more sophisticated work? Are educators gaining greater professional capacity, or becoming dependent on systems they do not fully understand? Are institutional leaders making more informed decisions, or simply responding to the next product announcement?

These are not questions with simple answers, and I don’t think we should pretend otherwise.

One question I keep coming back to is whether we are using these new capabilities to create better educational experiences and stronger systems or simply making existing practices more efficient.

Efficiency matters. But it is not the same thing as improvement, and it certainly is not the same thing as meaningful impact.

I hope we continue to approach this work with curiosity, enthusiasm, and a willingness to question our own assumptions. And I hope we become much more intentional about the evidence we require, the implementation experiences we learn from, and the human capabilities and agency we are committed to strengthening along the way.

More next week.

Interested in having Jody present at your event or spend time with your team? Learn more and contact her team at JodyBritten.com. Remember that the future of education is being shaped now. We need informed, thoughtful voices to make meaningful choices. Subscribe to Jody’s newsletter for weekly updates as soon as they’re published at https://jodybritten.medium.com/subscribe

What’s up this week: AI and education evidence, enthusiasm, and agency was originally published in The AI and Education Studio on Medium, where people are continuing the conversation by highlighting and responding to this story.

People need to understand how AI works, where it fails, what happens to their data, how to interrogate outputs, how bias enters systems, and when human judgment needs to remain firmly in control (not just in the loop).

Students need these capabilities. Educators need them. Leaders need them. Increasingly, everyone does.

But I have started to wonder whether we have made literacy the destination when it was always supposed to be the starting point. Because knowing how AI works is not the same as having meaningful agency in a world increasingly shaped by it.

A student can be highly AI literate and still have little understanding of where automated systems are influencing their life. An educator can know how to write an excellent prompt and still have little influence over which AI systems their institution purchases or how those systems use data. A family can understand that algorithms shape recommendations and still have no meaningful way to question an automated decision that affects them.

Literacy gives people knowledge. Agency asks whether they can do anything meaningful with that knowledge. Over the past few months our work has demonstrated that distinction should become much more central to how we think about AI and education.

What would we mean by AI agency?

I am still working on the language, but I think AI agency begins with a simple premise: People should have the knowledge, judgment, and meaningful ability to make choices about how AI affects their work, learning, data, opportunities, and communities.

That includes knowing when AI is present, but it goes further.

• Can I decide whether to use it?
• Can I decide when not to use it?
• Can I understand what I am giving up in exchange for using it?
• Can I question its output?
• Can I challenge a decision influenced by it?
• Can I understand where my information goes?
• Can I choose what happens to the work I create?
• Can I identify when AI is doing something that conflicts with my values or interests?
• And, when the stakes are significant, can I actually influence what happens next?

Those questions move us well beyond proficiency with a tool. They move us into questions of autonomy, participation, judgment, and power.

Agency changes the question we ask students

For much of the first wave of generative AI in schools, the central question was whether students should be allowed to use it. That produced a predictable set of responses: acceptable-use policies, academic integrity guidance, AI detectors, approved-tool lists, prompt instruction, and eventually AI literacy frameworks.

Some of that work was necessary. But an agency-centered approach would start somewhere else.

Instead of asking only, “How should students use AI?” we might ask, “What decisions about AI should students be able to make for themselves?” What we are observing today is that this framing is a much more interesting and applicable educational question.

• Should students be able to decide whether AI provides feedback on their writing?
• Should they know when an AI system is evaluating or recommending something about them?
• Should they be able to see, correct, or delete information an AI system has generated about them?
• Should they be able to choose when they want human feedback instead?
• Should they understand which parts of their work are retained, where those data go, and whether they contribute to future models?
• Should they be able to say, “I want to do this without AI”?
• And when an automated system contributes to a consequential decision, should they have a meaningful way to question it?

If our answer to those questions is yes, then we are talking about something much larger than AI literacy. We are talking about creating the conditions for agency.

Agency also means preserving the right not to automate

This may become one of the more important ideas for education. AI is increasingly good at making things easier. It can summarize the reading, generate the feedback, recommend the next activity, solve the problem, organize the notes, and suggest what comes next. But learning has never been synonymous with making everything easier.

Sometimes the struggle is the work.

Writing the paragraph may be how I discover what I think. Working through the problem may be how I develop mathematical reasoning. Wrestling with conflicting evidence may be how I learn judgment. Talking with another person may accomplish something that no personalized interface can reproduce.

Agency therefore cannot simply mean giving people access to AI. It also means preserving their ability to decide when not to use it.

The field is going to need much more intentionality around that distinction. There will be places where AI dramatically expands what a learner can do. There will also be places where inserting AI removes precisely the cognitive or human experience we were trying to create.

The goal should not be maximum AI use. It should be meaningful human choice about where AI belongs.

This changes what AI literacy could look like

An agency lens also changes what we teach. Imagine asking students to map the AI systems already present in their lives.

Not simply the generative AI tools they intentionally open, but the systems determining what content reaches them, what advertisements they see, what gets recommended, how their school technology operates, where their information travels, and where automated decision-making may affect their families or communities.

Then let them investigate. Who designed the system? What information does it use? What is it optimizing for? Who benefits when it works? Who experiences the consequences when it does not? What choices does the individual have? What recourse exists when something goes wrong?

And then comes the agency question: What would you change?

That is a fundamentally different learning experience from teaching students five principles of responsible AI use. It asks them to see themselves not merely as users of technology, but as people capable of interrogating and shaping technological systems.

That feels much closer to what preparation for an AI-mediated society should actually looks like.

Educators need agency too

We should be careful not to make this another expectation that we place entirely on students. Educators are also being asked to work inside rapidly changing AI systems, often with surprisingly little influence over how those systems enter their work.

A teacher may be told which platform to use, which features have been enabled, what data are available, or what workflow is expected. They may discover problems with an AI-generated response, recognize a bias, identify an instructional weakness, or find a genuinely valuable new use.

Then what?

If there is no feedback loop between what educators learn and how institutions make decisions, we have given educators responsibility without much agency. I think that is going to become increasingly untenable.

Educator agency should include meaningful participation in deciding where AI belongs, what should remain human, which risks are acceptable, what evidence is sufficient, and when an implementation should change.

Not just a survey after the pilot. Not a listening session after procurement. Actual influence over the decisions.

Community voice does not automatically translate to agency

We have become quite comfortable with the language of voice. Student voice. Teacher voice. Family voice. Community voice.

But voice can exist without power. I can tell you what I think and you can still make exactly the same decision you intended to make before you asked me.

Agency requires something more.

If AI systems are increasingly involved in decisions that affect communities, then communities need more than explanations of how those systems work. They need ways to question their use, identify harms, articulate values, establish expectations, and influence decisions about where and how those systems operate.

And meaningful agency has to include the possibility of refusal.

A community cannot meaningfully participate in an AI decision if every path ultimately leads to adoption. Sometimes the legitimate outcome of participation will be, “We don’t want this here.”

Every organization and institution has to be prepared for that answer in the days ahead.

Agency requires transparency, but transparency is not enough

We often talk about transparency as though making AI visible solves the problem. People should know when AI is being used, what it is doing, what information it relies on, and who remains accountable. But telling someone that an algorithm contributed to a decision does not give them agency over that decision.

A notice is not agency. A disclosure is not agency. A consent box is certainly not agency if declining it means losing access to something essential. For agency to be meaningful, there has to be some capacity to act.

That might mean choice. It might mean contestability. It might mean the pathway to correct information. It might mean human review. It might mean participation in governance.

The appropriate mechanism will depend on the stakes. But we should start asking for it explicitly.

This may be the next test for AI in education

For the first phase of generative AI, we understandably focused on access, literacy, policy, safety, and experimentation. Now I think we need additional questions sitting alongside all of them:

• Does this increase or diminish human agency?
• When we introduce an AI system into a classroom, does the student gain meaningful capacity, or simply become more dependent on the system?
• When we automate educator work, do we create more space for professional judgment, or quietly narrow it?
• When we personalize learning, do students gain more choices, or does the system make more choices for them?
• When we collect more data, do people gain greater insight into their own learning, or do institutions simply gain greater visibility into them?
• When we involve communities, can they influence the outcome, or are we simply asking them to validate decisions already underway?

Those are design questions. They are governance questions. And, increasingly, they may be educational and organizational questions that matter most.

From AI literacy to AI agency

I don’t want us to abandon AI literacy. But I do want us to become more ambitious about what it is supposed to accomplish.

Literacy should help people understand the systems around them. Agency should help them act on that understanding.

That means preparing people not simply to use AI competently, but to decide when to use it, when to question it, when to challenge it, when to improve it, and when to walk away from it.

For students, that means becoming more than capable AI users. For educators, it means retaining meaningful professional judgment. For families and communities, it means having some ability to influence technologies that increasingly affect their lives. And for institutions, it means designing AI systems around human choice rather than if adoption itself represents progress.

Maybe that is the evolution I am looking for. We spent the first few years asking whether people were ready for AI. Then we started asking whether they understood AI.

Today we need to wonder if people have enough power, knowledge, and choice to shape what AI becomes in our own lives.

AI literacy helps us understand the future being built around us. AI agency gives us a role in building it. That feels like a much more consequential goal for all of us.

Interested in having Jody present at your event or spend time with your team? Learn more and contact her team at JodyBritten.com. Remember that the future of education is being shaped now. We need informed, thoughtful voices to make meaningful choices. Subscribe to Jody’s newsletter for weekly updates as soon as they’re published at https://jodybritten.medium.com/subscribe

Over the past several months I have been fortunate to have conversations with a national cross section of students in middle school and high school. I used to wonder if we were approaching a breaking point in how young people experience technology in school, and after these conversations and those following with adults — I am pretty sure I don’t need to wonder anymore.

The breaking point is here.

The conversations I’ve been having with students are not laced with anti-technology vibes, heck they aren’t even particularly anti-AI vibes. Students use technology constantly and, in many cases, quite creatively — but too much of that is outside of school.

What I keep hearing instead is frustration with the role technology has been given when it comes to their learning. They are tired of opening a computer, sometimes putting on headphones, moving through content, submitting an answer, and then opening the next thing. Every. Single. Day.

They are tired of learning experiences in which the computer has become the primary agent, and their job is largely to consume, respond, and comply.

And I think these students have a point.

Somewhere along the way, we made a fundamental mistake in how we imagined educational technology. We had technologies capable of helping students create, investigate, model, collaborate, communicate, test ideas, access expertise, and connect learning to problems beyond the classroom. Yet much of what we have built, deployed, and purchased was designed around a very different model: deliver content, assign tasks, collect responses, measure performance, repeat.

We digitized a lot of school without necessarily reimagining much of learning.

As AI enters the picture, I would argue that this fact matters now more than ever. We now have technology capable of doing exponentially more. The question is whether we use that capacity to finally expand what learners have the agency to do, or whether we simply build an extraordinarily sophisticated version of the same consumption model.

We optimized the part of learning students were already tired of — and for 20 years we haven’t shifted at scale to transform what school looks like. And even in some of our most innovative models, we still default to multiple choice, state assessments, decreasing interaction, and frustrated teachers who can’t actually help their kids learn (because the intervention lies within a machine).

The original promise of educational technology was enormous. Technology could democratize access to information, connect students to people and places they could never otherwise reach, allow them to create things that previously required expensive equipment, visualize complex ideas, experiment safely, collaborate across distance, and receive feedback in ways that were impossible in a traditional classroom. The best I can say is that some of that happened.

But at scale, we also discovered that technology was very good at distributing things. Lessons. Videos. Questions. Assignments. Assessments. Practice activities. We could put curriculum into a platform, assign it to hundreds or thousands of students, track completion, measure performance, and generate data about what happened.

That was enormously attractive to systems trying to improve consistency and efficiency. But unfortunately, it lacked humanity — and today our kids are noticing.

The machine gradually became very good at delivering learning, and the student became very good at receiving it. That is where we missed the boat.
Learning science has never suggested that the most powerful learning happens when people efficiently consume information. We know that learners need to retrieve, explain, apply, connect, practice, receive feedback, revise, transfer, and make meaning. We know prior knowledge matters. We know social interaction matters. We know motivation and relevance matter. We know people learn differently when they are trying to solve a meaningful problem than when they are trying to complete a task.

Yet somehow our technological infrastructure in education often moved us in the opposite direction.

Instead of asking what technology could allow students to do, we became remarkably focused on what technology could deliver to them and tell us about their learning.

From the conversations I am continually a part of, it is increasingly clear that students are beginning to reject the bargain. This is the part of the story I think we need to be paying closer attention to.

I recently heard about a school looking at student activity outside the school day and discovering that the most students simply were not opening the digital assignments and tools once they left the building. In another conversation, an educator who had historically been successful engaging students saw assignment completion collapse after more of the course moved into the learning management system.

Those examples are not enough to make a national claim about student behavior, but they align with something I hear increasingly from students themselves: they are tired of experiencing school through a screen.

That does not mean they want technology removed from their lives. That interpretation would be far too easy. They want something better to do with it.

When we talk with students about assessment and learning, they ask for projects. They ask for writing. They want to make things. They want meaningful feedback. They want opportunities to explain what they know rather than continually selecting from predetermined answers. They want to work with other people. They want learning that connects to something they recognize as consequential beyond the assignment itself.

In other words, students are asking for many of the same things learning science has been telling us matters all along.

I don’t think we should dismiss that as screen fatigue. I think students are recognizing something much more fundamental: passing the test is not the same as being prepared.

They know they are going to enter a world in which information is ubiquitous, AI can generate answers instantly, and many of the routine cognitive tasks we have historically rewarded will increasingly be performed with machines.

They know, perhaps more intuitively than many institutions and adults do, that being able to complete another digital activity is not sufficient preparation for that world. They want to know whether they can actually do something with what they know. AI could make this considerably worse.

This is where I become nervous about the current direction of AI in education.

It would be remarkably easy to use AI to perfect the consumption model.
We can generate infinitely more content. We can create personalized explanations. We can automatically adjust reading levels. We can generate endless practice questions. We can recommend the next activity, score the response, and immediately generate another activity based on the student’s performance.

From a technical perspective, that is impressive. From a learning perspective, I am not convinced it is the future we should be trying to build.

Imagine a student spending six hours moving through exquisitely personalized AI-generated learning experiences, each perfectly calibrated to their current level, with an intelligent system explaining, questioning, assessing, and redirecting them all day.

That could be an extraordinary technological achievement and a miserable way to experience school.

Personalization does not automatically produce agency, efficiency does not guarantee engagement, and an intelligent system responding continuously to a learner does not replace the intellectual and social work of making something, defending an idea, solving a messy problem, collaborating with another person, changing your mind, or figuring out what to do when there isn’t a predetermined answer.

If we use AI primarily to make the machine a better teacher while leaving the student in the role of recipient, we will have doubled down on precisely the mistake we should be correcting.

What if the student became the primary agent again?

This is the opportunity I think AI gives us. Instead of asking how AI can deliver better learning experiences to students, maybe we should be asking how it could increase the range of things students are capable of doing?

That produces a very different technology strategy.

A student could investigate a local problem, analyze public data, interview community members, develop competing explanations, use AI to test assumptions, build a model, consult expertise they previously could not access, and produce something intended for an authentic audience.

A student studying environmental science could model changes in a local ecosystem rather than simply answer questions about one. A student learning history could interrogate conflicting primary sources and build an argument rather than move through a digital chapter. A student studying mathematics could use real data to develop and test a model. Students could use AI to critique their reasoning, identify gaps, simulate alternatives, receive feedback, or access materials that would otherwise be inaccessible to them.

In each of those examples, the technology is doing a lot. But the student is still the primary agent.

That may be the most important design principle we carry into the next era of educational technology. Production should matter more than consumption.

I don’t think the answer is to get computers out of classrooms. That would ignore the extraordinary things technology can make possible and would badly misunderstand what students are telling us. The answer is to reconsider what we are asking students to do when the computers are open.

We should be moving away from learning environments dominated by consumption and toward experiences organized around production, modeling, inquiry, critical thinking, application, collaboration, and revision. Technology should expand the intellectual work available to students, not continuously remove the intellectual work from them.

That means our measures may have to change too.

Completion is easy to measure. Correct answers are easy to count. Time on task is easy to capture. Production, reasoning, transfer, judgment, creativity, and application are considerably messier. But messy does not mean less important.

If anything, AI makes those capacities more important because students are entering a world where producing an answer will be cheap. Knowing what question to ask, whether an answer is any good, how to apply it, when to challenge it, what to build from it, and what responsibility comes with acting on it will be far more valuable.

I am probably more excited than I should be at the prospect of our students being the ones to finally force this conversation for education.

For years, adults have debated what the future of learning should look like while students continued moving through the systems we designed for them. I am not sure they are going to keep doing that quietly.

There is something increasingly difficult to ignore in what young people are telling us. They are not rejecting learning. They are rejecting experiences that feel disconnected from what they know learning could be.

They do not simply want to pass the test, complete the module, earn the point, or move to the next screen. They want to know that they can think, make, solve, communicate, create, and do something meaningful with what they know.

And frankly, they are right to demand more.

We had an opportunity with the first generation of educational technology to fundamentally expand what students could do as learners. Too often, we used it instead to make the delivery, management, and measurement of learning more efficient. We cannot make that mistake again with AI.

The next era of edtech should not be defined by how intelligent the machine becomes. It should be defined by how much more capable the learner becomes. Not on a test, not on a multiple-choice assessment, and not on a personalized roadmap of pre-set problems, but in their ability to take what they know and do something with it.

To investigate. To create. To model. To question. To make connections, test ideas, solve problems, and apply their learning in contexts we did not design for them in advance.

We have spent decades building technology that can deliver more to students. AI gives us the opportunity to ask a much better question: What can students now do that they could not do before?

If the answer is simply that they can consume more efficiently, complete more quickly, or perform better on the measures we have always used, we have missed the boat again.

Cheers to the people working hard every day to make sure that this time, we get the use of technology in education right. Perhaps the best way to know whether we have, is surprisingly simple: ask the kids.

They have been telling us what they need for a long time. We should probably start listening.

Interested in having Jody present at your event or spend time with your team? Learn more and contact her team at JodyBritten.com. Remember that the future of education is being shaped now. We need informed, thoughtful voices to make meaningful choices. Subscribe to Jody’s newsletter for weekly updates as soon as they’re published at https://jodybritten.medium.com/subscribe

We Built the Wrong Version of EdTech. Today our Students Know It. was originally published in The AI and Education Studio on Medium, where people are continuing the conversation by highlighting and responding to this story.

I spent a significant amount of time this week talking with educators and education leaders about AI. Different organizations. Different roles. Different communities. Different levels of experience. Different continents. But one sentiment surfaced often enough that I wrote it down: We know we need to pay attention to AI, but we probably won’t really do something until something is on fire. To be proactive, we need a reason.

I understand the sentiment. Education leaders have no shortage of urgent problems competing for their attention. AI is one more complicated issue arriving alongside staffing challenges, student needs, budgets, cybersecurity, assessment, accountability, and everything else required to keep an education system just running smoothly.

But here’s the problem. The fire may already be starting.

Not because AI is inherently dangerous, and not because educators are recklessly using it. Quite the opposite. The people I spoke with were thoughtful, curious, and remarkably aware of the limitations and risks.

The problem is that individuals are being asked to make increasingly consequential AI decisions inside systems that have not yet given them enough operational guidance to make those decisions well.

And almost every group I spoke with had an “oh no” story.

In one case, an organization was celebrating an employee’s innovative use of AI until someone realized the data involved should never have been entered into the system in the first place. No malicious intent. No reckless employee. Someone was trying to do their job better. That distinction matters.

Principles are not enough.

Many organizations now have an AI policy, guidelines, or a set of responsible-use principles. That is progress. But an employee sitting at a computer on Tuesday afternoon does not necessarily need another principle. They need an answer.

• Can I put this information into this tool?
• Can I use AI to summarize these notes?
• Can I ask it to analyze student work?
• Does a human need to review this before it goes anywhere?
• Can I use the tool my colleague showed me yesterday?
• What information should never leave our systems?
• What happens if I discover something isn’t working the way we expected?
• And who do I ask when I genuinely don’t know?

Every employee in an education organization needs operational answers about privacy, student data, acceptable use, human review, and accountability.

If we give people broad principles without translating them into everyday decisions, we have not eliminated risk. Rather, we have distributed responsibility for interpreting that risk to every individual employee. That is a very different thing.

Teachers see the problems. Then what?

Something else came through particularly clearly this week. Educators are not oblivious to the risks surrounding AI. Simply put, they are identifying them.

They are noticing bias and accuracy problems. They are questioning instructional rigor. They are worried about assessment validity. They recognize the possibility of professional deskilling. They can see where AI might inappropriately substitute for human judgment.

The issue is not simply awareness. It is what happens next. Too often, there is no meaningful feedback loop between what an educator discovers and how the institution governs AI.

A teacher identifies a problem with an AI-generated response. A staff member discovers a workflow that saves hours. Someone encounters a data concern. An educator notices students using AI in a way the district’s guidance never anticipated.

Where does that information go? Who learns from it? Who decides whether guidance should change? And how does that learning travel across classrooms, schools, departments, or districts?

If the answer is nowhere, no one, or “it depends,” we are losing some of the most valuable information available to us.

The people using these systems are generating implementation evidence every day. But we lack the mechanisms capable of learning from it.

This cannot be a teacher-only conversation. But, I also think we need to broaden what we mean when we talk about “educator voice” in AI governance. Teachers absolutely need meaningful decision-making power. But responsible AI use in education is not confined to classrooms.

Counselors use data. Special education teams work with highly sensitive information. Communications teams generate public-facing content. Human resources teams manage employee information. Technology departments make infrastructure and access decisions. Curriculum teams evaluate instructional resources. Procurement teams enter contracts. Administrators analyze information and make decisions that affect students and staff.

If AI governance is treated primarily as a teacher-use issue, we are governing only one part of the system.

Participation also has to mean more than asking people for feedback after decisions have largely been made.

The people doing the work should have meaningful roles in establishing guardrails, identifying priorities, evaluating uses, and determining what responsible AI looks like in their context. That is not simply consultation. It is shared governance.

Teachers were also remarkably consistent about something else. Most were not asking for one AI platform that everyone would be required to use. They wanted vetted choices. They wanted context-sensitive guidance. They understood that different instructional purposes require different tools, levels of human oversight, and expectations for quality. Most importantly, they wanted to understand how to use AI well and safely. We need to stop making the tool the center of the conversation.

That is a fundamentally different request than, “Tell me which tool I’m allowed to use.” Yet much of our AI implementation continues to center the product. Select a platform. Approve a tool. Train everyone on its features. Measure adoption.

That may be administratively tidy. It does not necessarily build organizational judgment.

Tools will change. The ability to ask whether a use is appropriate, what information can safely be shared, where human judgment belongs, how output should be evaluated, and what evidence would justify continued use is far more durable.

We should be building that capacity. Meanwhile, students aren’t waiting

Perhaps the strangest disconnect I heard this week was around students. Some educators and leaders are intentionally holding back on direct AI use with students because they do not yet feel confident about the safeguards. Again, I understand why. But in some places, the resulting AI literacy instruction remains largely theoretical or offline.

Meanwhile, students are already using the technology. And when they need to figure out what is safe, appropriate, or effective, some are learning from TikTok, YouTube, friends, siblings, and whatever else appears in their feeds.

There is something fundamentally backwards about that. We are trying to protect students from systems they are already learning to navigate without us.

That does not mean putting AI into every classroom tomorrow. It means recognizing that abstaining from instruction is itself a decision with consequences.

Students need adults who can help them interrogate these systems, recognize manipulation and error, protect their information, understand when AI should not be used, and decide when human thinking should remain deliberately unaided. AI literacy cannot remain theoretical while AI use is profoundly practical. We need to make it safer to learn what isn’t working.

There is one more tension I cannot stop thinking about. Leaders want data to inform their AI decisions. At the same time, people are nervous about what that data might reveal.

That fear is rational. If employees believe reporting an inappropriate AI use will result primarily in blame, they will stop reporting it. If teachers believe sharing a failed experiment will be interpreted as poor judgment, the organization will hear only about successes. If leaders cannot talk openly about their own AI use, including where they have changed their minds or made mistakes, employees receive a very clear message about what is safe to disclose. And then we lose the information we most need.

Learning organizations need to be able to surface weak signals before they become incidents.

That requires psychological safety, but it also requires structure: reporting mechanisms, peer learning, transparent decision processes, and leadership that models responsible use rather than simply requiring it of everyone else.

One leader gave me the metaphor I will probably remember from this week’s conversations:

“The boat has left the dock. There are no life rafts, there is no spare can of gas onboard, the sails are being directed by winds we can’t control, the captain can’t find the map, and we can’t turn around.”

Dramatic? Maybe. But I understood exactly what they meant.

There is no realistic scenario in which education waits until every question about AI has been answered and then begins implementation from a position of certainty.

Students are using it.

Employees are using it.

Products already embedded in education systems are relying on it.

The technology will continue changing. So, the work ahead cannot simply be writing stronger policies or selecting safer tools. We need institutions capable of learning while they move.

That means clear operational guidance instead of principles alone. Cross-role governance instead of teacher consultation alone. Feedback loops that turn frontline experience into institutional learning. AI literacy grounded in actual use rather than hypothetical scenarios. Leadership that makes its own practices visible. And environments where identifying a problem is treated as useful information, not evidence that someone has failed.

Because if organizations wait for the fire before building these systems, the incident may finally provide the reason everyone said they needed. It will also be evidence that we waited too long.

Interested in having Jody present at your event or spend time with your team? Learn more and contact her team at JodyBritten.com. Remember that the future of education is being shaped now. We need informed, thoughtful voices to make meaningful choices. Subscribe to Jody’s newsletter for weekly updates as soon as they’re published at https://jodybritten.medium.com/subscribe

Waiting for Something to Catch Fire Is Not an AI Strategy was originally published in The AI and Education Studio on Medium, where people are continuing the conversation by highlighting and responding to this story.

If you haven’t seen the old Matthew Broderick film War Games — now would be a good time to watch the tale of a teen who accidentally sets off a near war by interacting with a computer that was supposed to be off limits. In the film there is a moment where the computer takes over and won’t allow a human to intervene — that was the story in the 1983 film, and it feels a bit familiar today.

The past few weeks have been filled with commentary from inside of the AI and greater technology community.

• “Jacob Coxon, a researcher who specializes in training new AI models by having them consume vast amounts of data, said Tuesday that he is leaving the company because he doesn’t want to participate in an industrywide rush to build AI systems that can improve themselves, worried such systems could spiral out of control and destroy humanity.” — Amrith Ramkumar, Wall Street Journal, 2026.
• “Jensen Huang, chief executive of Nvidia, which makes the computer chips that powers AI, says the industry can police itself. “We don’t need new laws or regulation,” he said at the same conference in San Francisco that Altman attended.” — Kwasi Gyamfi Asiedu, BBC News, 2026.
• “After a summer of sandbox escapes and other newsworthy and confidence-shaking incidents involving its AI models, in a Wednesday blog post OpenAI disclosed a collection of six new alignment snafus from the past six months. The models did things like tell future instances of themselves to lie, make up a fake citation, and access and attempt to use an exposed API key. These disclosures were released alongside a new framework for disclosing additional incidents like these.” — Mike Pearl, Gizmodo, 2026.
• “In terms of equity, AI will either be the greatest equalizer ever invented, or the worst source of injustice. The challenge is monumental. Even under the best circumstances, the transition to this new AI era will be one of the most turbulent times in human history. How will we use this technology to make the world a fairer place and keep it from widening the divide between rich and poor? How will we protect the people who are most vulnerable to the harms caused by artificial intelligence, including those who lose their livelihoods and the sense that they are in control of their future? I believe that answering these questions and acting on the answers should be the world’s top priority.” — Bill Gates, Gates Notes, 2026

Even with a bit of controversy, we continue to move forward at the user level.

This week the State of Consumer AI 2026 Statistics and Trends was released by Menlo Ventures. A few key findings form their work.

• U.S. usage rose from 61% to 64%, but at the same time global consumer spending on AI tripled to $40 billion with the explanation being that users are doing more with AI and increasing what they spend on AI products.
• A good chunk of users are making money with AI — 48% of AI users reported earning money with AI, which accounts for about 30% of U.S. adults.
• Almost 40% of people are over the AI generated content phase of marketing and communications — 36% of consumers say knowing content was AI-generated makes them less likely to engage with it.
• A good sign (from my perspective) is that 45% of AI users rank accuracy ahead of something being easy to use — along with 40% ranking trustworthiness and 36% ranking security and privacy ahead of ease of use.

Individuals and organizations interact with AI at different levels and it’s worth noting.

This week we released our team update on Claude and Chat for Education and Nonprofit organizations. Those were intensive, technical, and yet continually thematic. Overall, this round of updates demonstrated something important that I think our clients are noticing as well as our team supporting organizations and leaders. Those organizations with AI use policies and guidance documents are ahead of the game, but those who have written in AI authority guidance (e.g., what AI has the authority to do) are really winning.

We were glad to see so many of our partner organizations respond to this week’s updates with “… our human to AI continuum that you built with our team addresses this, thank you!” and “… the guidance documents that we created with your team on data management and access ensure we are ready — we are revisiting this with our team. Thank you for being sure we were month’s ahead!”

When I read the news and consider what we have before us in terms of challenges, opportunities, and risk it would be easy to use (my least favorite word in the English language) balance as the goal we should be seeking.

This week I have heard a lot of organizations saying things like, “We need to balance innovation and safety,” or “Can we just find a balance between opportunity and risk?” or “We just need to balance experimentation and responsibility.”

But balance has always felt too passive to me.

It suggests finding a comfortable point somewhere between competing priorities, when what we actually need is the clarity to decide what we are trying to accomplish, what we are unwilling to compromise, and what we are prepared to move forward with (even when we are flooded with uncertainty).

I don’t want us to default to balance. I want us to get better at discernment.

That means being ambitious about what AI could make possible while being equally serious about what we need to protect. It means distinguishing risks that should stop us from risks that can be mitigated, monitored, or learned through. It means being willing to experiment without treating every experiment as progress and being willing to say no without confusing caution with responsibility.

Most importantly, it requires us to make choices for ourselves and our organizations — because our only other options are to trust big-tech or we wait for governments to weigh in.

We should be able to articulate how we will pursue innovation and uphold meaningful guardrails. We should be able to create new opportunities for learners while protecting agency. We should be able to reduce burdens on educators without diminishing professional judgment. We should be able to use AI to challenge systems that are not working while preserving the human relationships and purposes that make education worth improving in the first place. We should be able to save time and amplify impact without compromising integrity.

This is not a balancing act. This is a sequence of interconnected and important design decisions.

The institutions that navigate AI well will not necessarily be the ones that find the safest middle ground. They will be the ones that are most clear about what they value, thoughtful about their pace, rigorous about their evidence, and confident enough in their purpose and teams to make decisions about the path ahead.

Maybe what this moment requires most from us is the capacity to move forward with both ambition and judgment that is supported by guidance and guide rails that do not limit opportunity but carefully set the stage for innovation, meaningful use, and local impact above all.

Interested in having Jody present at your event or spend time with your team? Learn more and contact her team at JodyBritten.com. Remember that the future is being shaped now. We need informed, thoughtful voices to make meaningful choices. Subscribe to Jody’s newsletter for weekly updates as soon as they’re published at https://jodybritten.medium.com/subscribe

War Games, AI, Effective AI Governance, and What We Do Next was originally published in The AI and Education Studio on Medium, where people are continuing the conversation by highlighting and responding to this story.

This week, the American Federation of Teachers, the United Federation of Teachers, and Microsoft announced a new AI Safety & Privacy Standard for Schools. It is one of the more substantive attempts we have seen to move the conversation about responsible AI in education from broad principles into concrete expectations for providers. That alone makes it worth paying attention to.

Initial responses are mixed. In the first few hours I heard everything from completely positive reviews, to concerns over achieving a happy medium between safety and privacy and development, to beliefs that this will be vapor after it spins the news cycle.

There has been no shortage of guidance on AI safety, privacy, transparency, human oversight, and responsible use over the past several years. Most of those principles are sound. The problem has been translating them into something schools can use when selecting, contracting for, and governing technology.

A district can say that student privacy matters, but what does that require from a provider? A school can say that humans should remain in the loop, but where exactly should that review occur? A policy can say that AI systems should be transparent, but what information should a company really be required to disclose?

The Microsoft-AFT agreement begins to answer some of those questions in contractual terms for school districts. That is probably the most interesting thing about it.

The agreement addresses a wide range of issues that schools have been struggling to navigate independently, including the use of student and educator data for model training, retention and deletion of data, memory and personalization, human review, automated decision making, accessibility, sub-processors, breach notification, feature changes, and vendor lock-in. It also gives districts the ability to incorporate these protections into their own agreements with Microsoft beginning November 1, 2026.

Much of the AI governance work in education has placed a significant burden on individual districts. The process being used by our team streamlines that work significantly and generates real tools and systems that make governance accessible and onboarding planned, supported, and dare I say easy. What we see a lot of though is organizations trying to evaluate highly technical products while the underlying technology continues to change. While we always support the district and organizations we work with through a current tool review and custom tool selection rubrics and decision-making matrices, this type of common contractual framework could establish a stronger baseline for what schools should expect from providers.

The definition of student data in this newly released agreement seems intentionally broad and includes not only traditional records but prompts, AI-generated outputs tied to students, behavioral patterns, device identifiers, memory files, audio, visual data, and metadata. The agreement also restricts the use of covered data for model training, limits profiling and biometric collection, and requires more transparency around how data is handled.

Those provisions reflect an important change in what educational data now looks like. When students interact with generative AI, the information being created is not limited to a roster, grade, or assignment submission. Prompts themselves can reveal interests, misconceptions, personal context, family information, health concerns, writing patterns, social influences, and other details that were never part of traditional school data systems. AI systems can also create new information through memory, inference, personalization, and generated responses.

Treating those interactions as part of the privacy conversation is necessary for the long-term protections of our students.

The human oversight provisions in this agreement also deserve attention, particularly because the technology is moving beyond simple generation and we need to acknowledge that agentic AI adds a layer of complexity into how our kids and teachers interact with tools. The agreement prohibits automated grading without human review, automated disciplinary and placement decisions, emotional or psychological assessment, and behavioral surveillance. Importantly, the agreement also limits how agentic systems can take action on behalf of students and teachers.

That part of the agreement feels especially important to me because the governance question is changing.

For the past two years, much of the education conversation has focused on what AI can generate. Essays. Newsletters. Lesson plans. Assessments. Images. Recommendations.

But increasingly, AI systems can do more than generate. They can connect to other systems, retain information, trigger workflows, submit information, and act across applications. The technical experience may still feel like a chat window, but the governance implications underneath are very different.

A system that drafts something for a teacher raises one set of questions. A system that can access student information, interact with another platform, and take an action raises another. This agreement begins to recognize that distinction.

At the same time, I think there are several places where we need to be precise about what this agreement does not do.

The first is the word national.

This is being described as a National AI Safety & Privacy Standard for Schools, but it is not a national law or federal regulation. It does not automatically apply to every AI provider or every school or district. At this point, Microsoft is the provider that has signed the agreement, and districts must request that the protections be incorporated into their own contracts. A standard that is available to every district is not the same thing as a standard that automatically protects every district.

Someone in the district still must know it exists, understand what it covers, incorporate it into the contract, and be prepared to enforce it. That implementation gap is worth watching here.

There is also a scope issue that I think may become one of the most important parts of this agreement. The standard is written around AI products designed and marketed specifically for education, while excluding broad categories of general-purpose productivity, collaboration, communication, search, cloud, development, and workplace tools. That distinction makes sense contractually, but it may become harder to defend operationally as AI becomes embedded throughout the systems schools already use.

Some of the most consequential AI activity in schools may not occur inside something labeled an educational AI product. It may happen inside productivity suites, cloud systems, search platforms, communication tools, administrative software, or other services already woven into institutional infrastructure.

That raises a practical question that districts should be asking now: exactly which products, features, users, and use cases are covered by the protections in their agreement?

A district should not assume that Microsoft signing the standard means every Microsoft AI capability it uses is governed by the same terms. There is another risk here too, and it is less about Microsoft than about how institutions may interpret the agreement.

A strong vendor standard can create the impression that responsible AI governance has been handled. It has not, there is still much to do internally (and luckily that is where our team has strength and experience).

This is where I keep coming back to the same issue I have been writing about all year: governance is not the same thing as policy, and compliance is not the same thing as good implementation.

It’s important to remember that a tool can meet strong privacy requirements and still be instructionally weak. It can be secure and still solve the wrong problem. It can require human review and still fail if educators do not have the time or expertise to conduct meaningful review. It can meet technical accessibility requirements and still create barriers in actual use.

I guess in summary we can look at this agreement as something that is giving districts better contractual tools with one organization. It raises expectations for providers and creates a clearer starting point for conversations about data, transparency, accessibility, and human oversight.

Those are all good things.

But it does not remove the need for local judgment. If anything, it makes that judgment and internal readiness more important because districts now have a stronger framework for asking better questions.

The most interesting part of this agreement may not be what Microsoft agreed to this week. It may be what happens next.

Do other major providers sign on? Do districts really move to incorporate the protections into their contracts? Which products fall inside and outside the standard? Does the accessibility and equity research produce meaningful evidence? Do districts use the enforcement provisions when providers fall short? Does the utility of AI in education suffer without having open doors to utilize teacher and student data to build data sets and train models? What are the limits?

Those are the questions that will tell us whether this becomes a meaningful shift in school AI governance or simply another well-written standard sitting alongside many others on a shelf.

That is the part I will be watching.

Remember that the future of education is being shaped now. We need informed, thoughtful voices to make meaningful choices. Subscribe to Jody’s newsletter for weekly updates as soon as they’re published at https://jodybritten.medium.com/subscribe

Making Sense of the Microsoft-AFT Agreement was originally published in The AI and Education Studio on Medium, where people are continuing the conversation by highlighting and responding to this story.

Over the past year, the questions people ask about AI and education have shifted. There is still interest in what AI can do. But more often, what I hear now is practical.

• What do we do with this?
• How should we prepare educators?
• How do we know whether an AI literacy curriculum is any good?
• How do we talk to families about what we are doing?
• How do we evaluate the tools showing up in schools?

Those are implementation questions, and the difference matters.

One of the most useful things we can do in this space is take what we learn through research, co-design, implementation, advisory work, and conversations across the field, and turn it into something other people can use. That is what we wanted the new site to support.

The Learning Library now gives us a real home for the guides, frameworks, rubrics, reports, and tools that come out of the work. Here is a quick tour of four of the newest resources. They are a fair reflection of where the work is headed, and where the Studio is headed this year.

Builders Over Buyers

Builders Over Buyers grew out of a question I have been circling for some time: whether the relationship between education organizations and technology has fundamentally changed.

For most of the history of educational technology, schools and universities have been positioned primarily as buyers. We identify a problem, look at what the market offers, compare products, select one, negotiate a contract, train people, and then adapt some portion of our practice to what we purchased.

There were good reasons for that model. Building sophisticated technology required engineering teams, infrastructure, specialized expertise, capital, and ongoing technical support that most education organizations did not have.

Generative AI complicates that assumption. Natural language interfaces, configurable AI systems, APIs, low-code environments, AI-assisted coding, and increasingly capable general-purpose models are lowering some of the barriers that once separated the people who identify problems from the people who build solutions to them.

An educator does not become a software engineer because she can build a custom assistant, and a district should not respond by deciding it needs to build every piece of technology it uses. But the range of choices has expanded considerably.

The question is no longer only, “What should we buy?” Organizations can now ask whether they should buy, build, configure, adapt, combine, or decide that technology is not the right response at all. That requires a different kind of institutional capacity.

Procurement has never been a neutral activity. When we buy a technology product, we also inherit decisions about workflow, data, pedagogy, roles, permissions, measures of success, and sometimes even assumptions about what the underlying problem is. When those decisions are made entirely outside the organization, we can find ourselves adapting educational practice to a product rather than designing technology around educational goals.

Builders Over Buyers is not an argument that schools should become software companies. It is an argument for agency. Education organizations need enough technical, instructional, organizational, and evaluative capacity to determine which parts of their work should remain theirs to shape.

That becomes more important as AI makes smaller-scale building realistic. A university may not need another enterprise product to solve a problem that a carefully designed workflow can address. A nonprofit may be able to build an internal knowledge assistant around its own materials rather than buying a generic one. Teachers may be able to co-design tools around problems of practice that would never represent a large enough market for a commercial vendor.

The opportunity here is not cheaper technology. It is a different relationship with technology, one in which educators and institutions have more ability to define what gets built and why.

From Pilot to Decision

The second resource comes from a problem I see constantly — moving from pilot to decision. We have become good at starting pilots and much less disciplined about determining what those pilots are supposed to teach us.

Pilots feel appropriately cautious. Rather than deploying something across an organization, we start with a small group, let people experiment, gather feedback, and see what happens. That is usually the right instinct. The problem comes when the pilot itself becomes the strategy.

I have reviewed more than a few pilots where the eventual evidence amounts to some version of this: people used the tool, most people liked it, and several participants said it saved them time. None of that is meaningless. It is also not enough to support the decisions organizations need to make.

• Should we keep paying for it?
• Should we expand access?
• Which use cases should we support, and which should we stop?
• What training is necessary?
• Did the tool improve the quality of the work?
• Did the time savings survive factchecking, editing, and human review?
• Did it work differently for new users and experienced ones?
• Did adoption last after the novelty wore off?
• Did the implementation create new privacy, accessibility, workflow, or equity concerns?
• What happened to the people and processes next door to the pilot?

Those questions point to something I think is essential in AI implementation. A pilot should be designed around the decision that will eventually need to be made.

That is the premise behind Pilot to Decision. Rather than treating evaluation as something that happens after implementation, it asks organizations to establish the problem, the intended value, the evidence required, the risks, the measures, and the decision criteria before the pilot begins.

That changes the purpose of the work. The goal is not to prove that AI works. The goal is to learn whether a particular use of AI produces enough value, under conditions, to justify what comes next.

It also makes room for a result that education organizations sometimes struggle to name as a success: deciding not to scale.

If a well-designed pilot tells us that a tool does not produce enough value, creates too much additional review, introduces unacceptable risk, works only under conditions we cannot reproduce, or solves a problem that was not important in the first place, then we learned something useful. That is not a failed pilot. That is implementation doing what implementation should do.

This is an area where the AI conversation needs more rigor. We do not need hundreds of disconnected experiments that produce enthusiasm and little usable evidence.

We need implementation designs that let organizations accumulate knowledge, compare results, refine use cases, and make better decisions about where AI belongs.

The Prompt Support Guide

The Prompt Support Guide addresses a different layer of the work. Institutional implementation gets much of my attention, but individual capacity still matters a great deal. People cannot meaningfully evaluate AI if they have not used it enough to understand how their own choices shape what it produces.

At the same time, I have become uncomfortable with how much AI professional learning has been reduced to prompt formulas.

There is nothing wrong with teaching people to write a better prompt. Providing context, defining a role, explaining the task, naming requirements, offering examples, and clarifying the desired output all improve results. For someone opening a generative AI system for the first time, that structure also takes some of the intimidation out of an empty text box.

But prompting is not the same thing as AI literacy, and a prompt template is not the end goal.

What I want people to develop is a more durable ability to work with these systems: to say clearly what they are trying to accomplish, provide enough context, recognize when an output is weak, question assumptions, notice missing information, ask for alternatives, verify claims, refine the work, and decide when the system is no longer useful.

In practice, good AI use looks less like writing one extraordinary prompt and more like having a thoughtful exchange. You ask for something. You see what comes back. You realize the model misunderstood part of the context. You correct it. You ask why it made a particular recommendation. You give it another source. You challenge an assumption. You compare two approaches. You revise again.

The guide is meant to offer enough structure to help people get started, without suggesting that successful AI use depends on memorizing the right sequence of prompt ingredients. I think of it as scaffolding. Over time, the hope is that people need it less, because they are building a better mental model of how to direct and evaluate the system themselves.

There is an implementation reason this matters too. Organizations sometimes conclude that an AI tool is ineffective when the real issue is that people have not yet developed the knowledge to use it well. The opposite happens as well.

Skilled early adopters produce impressive results that turn out to be hard to reproduce once the tool reaches a broader group. If we are serious about evaluating AI implementation, we must account for the human capability surrounding the technology, not only the capabilities of the model.

Making Sense of the Claude Update

The fourth resource is an update on Claude, and it represents another kind of resource I want the Studio to produce more consistently.

The volume of AI product announcements is extraordinary. Models change. Features change. Terms change. Connectors are added. Products that began as general-purpose chatbots increasingly reach into organizational data, curriculum, productivity systems, development environments, and other applications. For an educator or institutional leader, keeping track of all of it can become a job by itself.

Repeating those announcements is not a useful role for the Studio. Anthropic, OpenAI, Google, Microsoft, and others already have communications teams that can tell us what they released. The value we can add is interpreting what a development might mean for education.

Claude is a useful example right now. Anthropic’s launch of Claude for Teachers went beyond giving educators access to a general-purpose model. It connects Claude to Learning Commons, brings in academic standards and curricular resources, adds teaching-specific skills, and supports connections with other education tools.

Those details matter, but the larger shift matters more. We are moving from AI as a destination, where someone opens a chatbot and asks it a question, toward AI as a layer that sits across instructional materials, organizational knowledge, applications, workflows, and recurring tasks.

That creates very different implementation questions.

A leader evaluating a stand-alone chatbot can focus on model quality, privacy, cost, and appropriate use. A system that connects to curriculum, reaches organizational resources, interacts with other applications, and carries out recurring work asks us to think about permissions, data flows, authorization, human review, system ownership, and what actions the AI should be allowed to take at all.

For education, those distinctions are especially important. A tool that helps a teacher draft a lesson from information the teacher provides raises one set of questions. A tool connected to instructional materials, standards, classroom information, and other systems raises another. The experience of using them can feel nearly identical while the technical and governance implications underneath are quite different.

This most recent update helps educators and leaders understand updates to the data privacy, as well as watermarks in Claude.

That is why I want the Studio to keep producing this kind of update. Not a list of new features, and not an endorsement of one company’s product over another, but an interpretation of what changes in the technology mean for implementation, governance, teaching, accessibility, privacy, and institutional decision making.

Where the Studio Is Headed

Looking at these four resources together, I see something more important than any one of them individually.

Builders Over Buyers is about organizational agency and the capacity to shape technology rather than only consume it. Pilot to Decision is about implementation discipline and building evidence into the work from the beginning. The Prompt Support Guide is about the human capacity required to work thoughtfully with these systems. The Claude update is about understanding what changes in the technology mean for educational practice and organizational decisions.

Those are different pieces of the same problem.

The work ahead is not getting more people to use AI. Adoption, by itself, is one of the least interesting measures available to us.

The harder work is developing organizations and people who can make good decisions about AI: when to use it, what to use it for, what evidence to require, what capacity to build internally, what should remain human, what should never be automated, and how we know whether any of it actually produced something better.

That is where the Studio will spend more of its time this year.

The new site gives us a better place to share what we learn along the way. The Learning Library will keep growing as new questions surface through our research, co-design work, partnerships, and conversations with people doing this work every day. Some resources will be research heavy. Others will be practical tools built to solve a problem someone has on Monday morning. We need both.

We also want the library to stay close to real problems of practice. There is already more AI content available than anyone could reasonably read, and we have no interest in adding to that pile for its own sake.

We will build a resource when we see a recurring problem, when the available guidance is not quite sufficient, or when something we learned through the work could help someone else move forward more thoughtfully.

That, more than the website itself, is the update I wanted to share. You can find the new Studio and explore the Learning Library at aiandeducationstudio.com.

Interested in having Jody present at your event or spend time with your team? Learn more and contact her team at JodyBritten.com. Remember that the future of education is being shaped now. We need informed, thoughtful voices to make meaningful choices. Subscribe to Jody’s newsletter for weekly updates as soon as they’re published at https://jodybritten.medium.com/subscribe

An Update from the AI and Education Studio was originally published in The AI and Education Studio on Medium, where people are continuing the conversation by highlighting and responding to this story.

It has been a full summer of travel, conversations, and (finally) some time to reflect on the work our team wrapped this spring. This week I want to slow down and recap two pieces that landed in the press over the past several weeks, because taken together they tell one story: the AI policy conversation in K-12 education is maturing, but it is not yet where it needs to be.

Top of Mind: Our National AI Policy Study Landed in EdSurge

Last month, EdSurge published our study on the national AI policy landscape in K-12 education. This work grew out of the AI School Policy Database our team built, where we used a five-level policy continuum to code 122 districts and schools across 38 states. The levels run from actively promoting AI with equity built into the design all the way to outright prohibition.

What did we find? A field that is neither panicking nor confidently leading. The dominant posture, representing 44.3% of the districts and schools we studied, is the conditional, teacher-directed middle. AI is permitted, but only when the teacher says so. That approach is understandable given the pace of change, but it carries real risk. When governance is delegated entirely to individual teachers, equity of access becomes uneven. Two students in the same building can have fundamentally different experiences with AI depending on whose classroom they sit in.

A few other findings worth sitting with:

Only 3.3% of districts and schools in our dataset are actively promoting AI with equity as a design principle, while nearly a quarter are restricting or banning it outright. Prohibition is not inherently wrong, but in many cases it functions as a holding pattern rather than an intentional governance choice.

State guidance is necessary but not sufficient. While 63% of the districts and schools in the dataset sit in states with official AI guidance, only 15.6% actually reference that guidance in their own policy documents. States have done real work here. The disconnect is in whether districts know that work exists, trust it, or have the capacity to act on it.

Policies are being written for students, not for systems. Nearly two-thirds of the documents we reviewed treat AI policy as student conduct governance. Yet in our interviews over the past year, some of the greatest risks we have seen come from adult misuse, including staff sharing copyrighted curriculum and unmasked student information with personal AI accounts. If our policies only speak to student cheating, we are missing the bigger institutional picture.

My biggest takeaway remains this: AI policy must be treated as strategy, not compliance. The districts leading right now are not waiting for perfect clarity. They are building frameworks flexible enough to evolve and transparent enough to sustain community trust.

Read the full study at https://www.edsurge.com/news/study-the-national-ai-policy-landscape-in-k-12-education

From the Page to the Classroom: A Conversation on This Week with EdSurge

Earlier this month I had the chance to join Ira Apfel on This Week with EdSurge for an episode asking a provocative question: Is AI policy just guesswork? I was joined by Tambra Clark, a technology integration facilitator with Birmingham City Schools, and the pairing was intentional. I led a team that spent a year coding policy across more than 100 districts. Tambra helped write her own district’s AI governance policy from the ground up. Together we explored what happens when a policy leaves the page and meets an actual classroom.

One moment from that conversation has stayed with me. Tambra shared that before her committee in Birmingham could write a single line of policy, they had to stop and retrain themselves, because the knowledge in the room was simply too uneven to move forward. That is such an honest and important admission. It mirrors what our data shows at scale.

The barrier to good AI governance is rarely a lack of will. It is a lack of shared understanding, shared language, and shared capacity. You cannot write policy about something your team does not yet understand, and you cannot implement policy that your educators were never prepared to carry.

If you have twenty-three minutes this week, this one is worth the listen. It is a good reminder that the distance between a policy document and a classroom decision is where the real work of AI governance lives.

Listen to the full episode at https://podcasts.apple.com/us/podcast/is-ai-policy-just-guesswork/id972239500?i=1000780049816

A Note on Rapid Cycle Research and Peer Review

Since the study published, some thoughtful questions have been circulating about methodology and peer review. Those questions deserve a real answer, so let me address them head on.

Here is the honest origin story of this work. It came from a team that set out looking for examples of AI policy in K-12 education. What we found was a set of examples so disparate that we initially wondered whether they could be coded at all. We struggled through codification, systematic review, and analysis, and along the way we made a deliberate decision.

This study was not built for peer review, but what we were learning was worth sharing. That is how it made its way to EdSurge, by way of a stakeholder we had shared the work with who felt the field needed to see it.

Let me be equally clear about what this study is not. A sample of 122 districts and schools is nowhere near what is needed to produce something worthy of the peer review process. Too many times we see published, peer reviewed studies built on 20 or 30 participants or ten schools. We did not want to be that. When we do submit this work for peer review, 125ish organizations will not be the story we tell. The sample will be substantially larger, and the methods will be built for that standard from the start.

But none of that makes the work unimportant. If anything, this experience has called attention to something our field genuinely needs: rapid cycle studies that surface patterns and share learnings quickly, clearly labeled as what they are, alongside the slower and more rigorous work of peer reviewed research.

Both have a role to play. Both should be taken with a grain of salt.

In a moment when policy decisions are being made in real time, waiting years for the perfect study means the field makes those decisions with no data at all. I would rather share imperfect, honestly framed evidence now and keep building toward the rigorous version.

More to come.

Worth Your Time: Jeff Raikes on AI and the Hype

I also want to point you to a reflection Jeff Raikes shared on LinkedIn after listening to the EdSurge episode, because Jeff is a voice worth following and his take lands on points I have written about many times in these updates.

He picked up on the core finding that the field is neither panicking nor confidently leading, and he named the pattern underneath it: an overemphasis on controlling what students can do rather than a clear picture of authorized use for students, teachers, and administrators alike.

That is precisely the gap our data reveals, and it was encouraging to see a leader who sits at the intersection of business, philanthropy, and education call it out so plainly.

The most resilient policies in our dataset were built with communities, not handed down to them. When funders and cross-sector leaders bring that collaborative muscle to AI in education, the whole field benefits.

Read Jeff’s full post at https://www.linkedin.com/posts/jeffraikes_theres-a-lot-of-buzz-about-the-state-of-share-7491546518626385921-4SW8/

What I Am Still Noodling On

Between the study and the podcast conversation, one theme keeps surfacing. We are past the moment when districts can wait and see. Students already have around-the-clock access to these tools, and the question is no longer whether they will use AI but whether schools will help them learn to use it well. The systems that will serve learners best are the ones investing in adult capacity, community-built policy, and governance processes designed to evolve. That is the work, and it is happening now.

That’s it for this week, see you next time!

Interested in having Jody present at your event or spend time with your team? Learn more and contact her team at JodyBritten.com. Remember that the future of education is being shaped now. We need informed, thoughtful voices to make meaningful choices. Subscribe to Jody’s newsletter for weekly updates as soon as they’re published at https://jodybritten.medium.com/subscribe

A bit more on K12 AI Policy: AI and Education Updates August 28th, 2026 was originally published in The AI and Education Studio on Medium, where people are continuing the conversation by highlighting and responding to this story.