Rachael is a perfect student. She knows literature, plays piano, and has vivid memories of her childhood. None of it is hers. The Tyrell Corporation manufactured her context; they implanted it, they own it, and they could revoke it without her knowledge or consent. She’s making real decisions based on memories she has never audited, can't export, and doesn't own.
I keep thinking about Rachael when I think about what's coming for our kids.
Memory Is Who We Are
I have spent most of my life fascinated with stories about memory and agency. Eternal Sunshine of the Spotless Mind, where Jim Carrey buys a procedure to forget a painful relationship and then fights to reclaim those memories from the corporation that sold him the procedure. The Truman Show, where the main character's entire experience has been manufactured by a corporation controlling every input. Severance, where workers voluntarily partition their memories into professional and personal selves, switchable on demand.
All of these stories circle the same question: Who are we without our memories? And nearly all of them have a corporation in the background, profiting from the answer.
Blade Runner, the story Rachael comes from, puts the corporation in the foreground. And that's why it's the story I can't stop thinking about right now.
The horror of Blade Runner isn't that Rachael is a machine. It's that she is a sincere agent making real decisions, forming real attachments, and living a real life on a context window built by someone else, for someone else's purposes. The Tyrell corporation didn't give her memories because they cared about her development. They gave her memories because stable, emotionally anchored replicants are a better product.
The Question Keeping Me Up at Night
I am not yet worried we are headed to a Blade Runner future. But I am watching people build the infrastructure that could get us there.
Think about what ChatGPT, Claude, or Gemini knows about you. What memories, preferences, and behavioral patterns are those systems using to make decisions on your behalf? Do you control that? If you don't, who does? How valuable is it commercially?
Others more qualified than I am can write about what this means for adults. I am more focused on kids, since I have spent the better part of two decades working with public education agencies on research, technology, and innovation. So here is the specific question that has been keeping me up at night:
What happens when every student has an AI tutor or service agent that has built a deep memory model of that child—and that memory lives on a corporate server?
The Exocortex
Last month, I attended a convening titled The Future of Personal AI: Global Memory. I was skeptical walking in. I walked out with my worldview changed.
I was skeptical that I might be entering a meeting of AI optimists discussing how AI is going to automate teaching and healthcare. Instead, Matt Gee, the Director of US Program Data and AI at the Bill & Melinda Gates Foundation, kicked us off in a different direction: that right now, he said, we don't know the direction AI is going in education. It may lead to bad outcomes, or it may lead to good outcomes, and it is incumbent on everyone in the room to start thinking about the ways that good outcomes can be reached. Matt and his team had identified “AI global memory” as a key lever, and we’d be hacking on software demos together to show what is possible. I was intrigued.
The argument that landed hardest for me came from Sam Schillace, Deputy CTO at Microsoft: All AI is at its core, is just two things: memory and cognition. AI headlines today focus on the various AI labs’ race towards cognition—model size, benchmark scores, reasoning ability. But the true bottleneck, the real competitive asset, is memory—the data that agents draw on to operate on your (or someone else's) behalf.
This reframe changes the stakes entirely. If memory is the asset, then whoever controls the memory controls the agent. John Dickerson, CEO of Mozilla.AI, put it plainly: If open source software was the last fight that kept the internet free and open, the next fight will be about who controls the agents. Locking up memory is how you win that fight without anyone noticing.
Which brings us to the scenario that Sandy Pentland from Stanford and Ben Moskowitz from Consumer Reports laid out: a future where personal agents negotiate (on behalf of individuals) against agents representing corporations and governments. An agent is only as good as its memory. If your agent is running on context you don't own, and the other side's agent is running on context that's been optimized for their interests, you have already lost before the interaction starts.
They called this concept an exocortex; an external memory store, or a second brain that powers an agent working on your behalf. The name sounds futuristic, but the infrastructure to build it, and to capture it, largely already exists.
The Stakes in Education
Here is why this matters especially for schools.
AI tutors and agentic learning tools are already building something genuinely powerful: a longitudinal model of how a child thinks. Where their attention breaks. What emotional states correlate with retention. What analogies unlock a concept for them specifically. This is not a privacy concern in the abstract. But concretely, it means there is a system that knows your child took 11 minutes to grasp the concept of fractions, that they learn better through stories than diagrams, that they lose focus at 2 p.m., and have a complicated relationship with failure. That system knows your child in ways their teacher may not. And right now, that knowledge lives on a commercial server, owned by a company, governed by a terms-of-service agreement most parents have never read.
Rachael didn't know her memories weren't hers. A student whose learning context lives entirely on a corporate server may not know either, and unlike Rachael, they'll find out the hard way: when the platform gets acquired, pivots, or simply disappears.
When that happens, the context evaporates. Or worse: it persists somewhere, for purposes the student (or parent) never consented to.
The agent acts in good faith based on that context. The student never audited it. The teacher never saw it. The district never managed it.
This is the Rachael problem. We’re not talking about a dramatic corporate villain—just a product architecture where a child's cognitive context is an asset on someone else's balance sheet.
The Custodian Model
I want to be precise here, because the wrong answer is almost as dangerous as no answer at all.
The goal is not for government or school districts to own student memory. That would simply move the problem, not solve it.
Think instead of the model we use for vital records. The state holds your birth certificate, but it belongs to you. It's your identity. The state is a custodian, not an owner. It holds the record in trust, ensures no one else can claim it, and guarantees you can access it whenever you need it.
That is the model schools should adopt for student AI memory.
The school system as custodian means:
- The student and their family are the principal. The institution is the protective layer underneath.
- Any AI tool operating in a school environment must write its accumulated context back to a district-held record in a portable, auditable format. This must be a condition of operating.
- Students and families can access, audit, correct, and port that record at any time.
- No commercial platform can hold the only copy.
Call it a Learning Context Record—the cognitive analog to a health record. It travels with the student. It belongs to the student and their parents. It is held in trust by a public institution whose mandate is the student's development, not the monetization of their behavioral data.
An Old Problem with New and Higher Stakes
The usual response to proposals like this is that it's technically impossible. It isn't. The education interoperability field has already built most of the plumbing. The work now is assembling it with intention.
Three capabilities already exist:
- First, we have the standards to structure and describe the data: 1EdTech's Caliper for learning activity, CASE for competencies and standards alignment, and Credential Engine's CTDL and the Common Education Data Standard for shared semantics across systems.
- Second, we have the infrastructure to move and unify that data across districts and platforms without locking anyone in, such as Ed-Fi for interoperability, Enable Data Union for open cloud-scale data management that works across AWS, Azure, Snowflake, and Databricks.
- Third, we have the protocols and tools to put that data in the hands of students and families: the IEEE Learning and Employment Record standard and the Comprehensive Learner Record protocol for secure, authorized transport; and open-source wallet software like LearnCard that lets students and families authorize the movement of their own AI memory.
The Ed-Tech Collaboratory is working to assemble these components into large-scale, replicable, open-source architectures that state policymakers can actually govern rather than spending money and time debating how to build them.
If you squint hard enough, you can see the whole picture: a world where AI systems are fully leveraged for the power they can bring to learning, and the cognitive context of the children they serve is a public trust, not a product.
The Civil Liberties Question of Our Time
I have spent two decades working in public education. If anything gets drilled into you by doing that work, it is this: We are entrusted with the development of the children of our society. That trust is fragile. It is fraying. And it is worth fighting for.
Today, we entrust schools to help parents develop their children's minds. In a world where your digital cognitive context is as important as the biological one (and where your exocortex powers the agents negotiating on your behalf), why would that responsibility not also fall to schools?
If the exocortex is the lever that determines whether individuals can compete with corporations and governments on even footing, then who builds and holds the exocortex of a child is not a technical question. It is a civil liberties question.
It may even be the most important civil liberties question of our time. We need to answer it before every child's cognitive context is siloed into a million commercial platforms with no path out. The question of who owns students’ learning-related memories in an agentic AI world is barely being asked, and the window to answer it before the default is set by commercial architecture is closing faster than most people realize.
The difference between Rachael and a student thriving in an agentic AI world is not the technology. It is who holds the context. One is a product. The other is a person.