Nineteen ninety-eight, two thousand six, two thousand thirteen, two thousand nineteen, two thousand twenty-six. The classroom robot has been replaced roughly every seven years for three decades. That is not bad luck and it is not anybody’s villainy. It is a schedule, kept by someone who is not in the room, and most schools have no instrument pointed at it.

I. A Date Somebody Else Already Chose

A school buys a cart of classroom robots. Somewhere in a building the school has never visited, a product manager who has never met them has already written down the year that cart stops being supported. The school does not know the year. It is not on the invoice, it is not in the catalogue, and nobody was hiding it — it simply never came up, because the transaction was about this year’s budget and the date belongs to a different conversation.

Six years later the date arrives. The software stops getting features. Then the competition the students enter stops accepting the platform. Then the district has an unbudgeted capital expense, a set of lesson plans that no longer match the screenshots, and a teacher who has to learn a new environment in a summer, on her own time, for the third time in her career.

Every part of that is ordinary. None of it involves anyone behaving badly. And it happens to schools over and over, in a rhythm regular enough to set a watch by, which is the part this essay is about.

II. The Cadence

Take the best-documented case in educational robotics and lay the dates out end to end.

Nineteen ninety-eight, two thousand six, two thousand thirteen, two thousand nineteen, two thousand twenty-six. The gaps are eight years, seven years, six years, seven years.

I want to be careful about what I am and am not saying. The new hardware may well be better than the old hardware; the successors usually have been. The company is entitled to build what it thinks is right, and a seven-year support life for a classroom electronics product is not stingy — it is longer than most consumer devices and longer than most school technology of any kind. The competition organizers running two editions in parallel through a transition are doing more than they had to do.

The observation is not that this is wrong. It is that it is regular, and that a regular event which a school treats as a surprise every single time is an instrument problem, not a vendor problem.

III. This Is Not About One Company

The same shape appears everywhere in education technology, and the examples are worth having in hand so that nobody mistakes this for a complaint about robots.

Flash. An entire generation of interactive courseware — simulations, tutorials, whole state-funded content libraries — was built on Adobe Flash. Support ended at the close of 2020 and much of that material simply stopped running. Some was rebuilt. A great deal was not, and the teachers who had spent years assembling it had no recourse at all.

Chromebooks, and the date on the platform rather than on the purchase. School Chromebooks carry an Auto Update Expiration date, after which the device stops receiving ChromeOS updates and security patches. Google extended the window substantially — many platforms now qualify for up to ten years of automatic updates, which is a genuinely good change and one schools asked for. But here is the mechanism a purchasing officer needs to see: the expiration is determined by the platform’s launch year, not by the date the school bought the device. A district that buys into a platform in its fourth year does not get ten years. It gets six, and the number was fixed before anyone at the district had heard of the model.

The Apple II. Apple donated machines into California classrooms in the eighties and the Apple II was the schoolhouse computer of its era. The line was discontinued in 1993. I designed the processor in it, so I am not describing someone else’s discontinuation.

The BBC Micro. An entire national computing curriculum in Britain was built around one machine from one manufacturer. When the machine went, so did the curriculum. The successor, decades later, was the micro:bit — and the interesting thing about that story is the governance choice: it was placed under a nonprofit foundation rather than a single company’s product line. Somebody had learned the lesson and moved the governor.

The pattern does not care what the technology is. Anything a school builds on, that a company owns, ends when the company decides it ends.

IV. The Exception, and What It Actually Proves

There is one famous counterexample in American classrooms, and it is worth examining closely because it looks like a happy story and is really a diagnostic one.

The TI-83 arrived in 1996. The TI-84 Plus followed in 2004. Thirty years later, that family is still sitting in high school mathematics classrooms, still being bought new, still doing what it did. Nothing else in educational technology has that record.

Why did it last? Not because Texas Instruments promised it would. It lasted because textbooks were written to it and, decisively, because standardized testing rules permitted it and prohibited the devices that had long since surpassed it. The governor that preserved that platform was not the vendor’s product roadmap. It was an examination policy, held by an entirely different organization, fixed years in advance, published, and not revisable by the manufacturer.

That is the Medium Separation (CKB-11), which this Foundation’s canon states for artificial intelligence and which turns out to describe a graphing calculator: a governor that lives in the same medium as the governed inherits that medium’s mutability, so durable constraint has to sit in a different substrate, fixed before the thing runs. A product roadmap governing a product is the same medium. An exam board governing a product is a different one — and thirty years of stability is what a different one buys you.

Now the honesty that keeps this from being a fable. That same arrangement is widely and fairly criticized as a lock-in that kept a generation of students paying for technology a phone had surpassed, and the durability and the stagnation came from the same mechanism. Stability is not the same thing as goodness. A governor in a different medium gives you durability whether or not what it is preserving deserves to be preserved.

V. What Is Actually Failing, in the Framework’s Terms

The Theory of Embedded Intelligence treats any system that senses, processes, communicates, and actuates as an intelligence, at any scale — a cell, a person, a team, a company, a school district. Run a district through that definition during a platform transition and the failure locates itself immediately.

The district’s Sense phase is well developed and points at exactly the things a district is built to notice: enrollment, budget cycles, test scores, teacher complaints, state requirements, the condition of the carts. Those all have addresses. Somebody’s job is to watch each of them.

The vendor’s product roadmap has no address in that apparatus. No role in the district is responsible for sensing it. It is not a line in the budget system, it does not generate a report, and it does not arrive as a complaint until the year it becomes an emergency. So the district cannot process what it has not sensed, cannot communicate a plan it has not processed, and cannot actuate a budget for a plan it has not communicated. The cycle breaks at the first phase, and everything downstream is not a failure of competence. It is a failure of aperture.

The previous essay in this series made the point about hardware: a phenomenon with no address cannot enter the Process phase, no matter how real it is. A slot gives a phenomenon an address. The same move is available here, and it costs nothing but the discipline of asking.

VI. Giving the Roadmap an Address

Here is the practical payload of this essay: six questions to ask a supplier before a purchase, and to get answered in writing. They are not adversarial, and a good supplier will answer them readily — several publish the answers already. A supplier who cannot answer them has told you something important at no cost to you.

Then do the thing that actually converts the answers into sensing: write the dates into the same calendar that holds the budget cycle, three years ahead of each one. A date that exists only in a vendor’s document is still a phenomenon with no address inside your institution. A date in your own budget calendar has one.

That is the whole intervention. It is unglamorous, it requires no technology, and it would have prevented most of what I have watched happen to schools over forty years.

VII. Writing a Unit That Survives Its Hardware

The second half of the answer is about what a curriculum is made of, and it is where the Foundation has actually had to change how it writes things.

When a platform is retired, some things in a lesson die and some do not, and the difference is predictable.

So the Foundation’s project sheets are written against functions rather than products: a sensor readable over a bus, a decision made in code or in logic, a path that carries that decision, an output that changes something. Each sheet names at least three unrelated suppliers whose parts satisfy the function, and every unit has to be runnable with no purchase at all. That rule was written for a different reason — to keep my own commercial interest off the material — and it turns out to solve this problem too, which is usually a sign that a rule is load-bearing rather than decorative.

The test is simple and a curriculum author can apply it in an afternoon. Take any unit you have written. Assume the hardware in it is discontinued tomorrow. How much of the document has to be rewritten? If the answer is most of it, you did not write a curriculum. You wrote a manual for a product, and it will expire on that product’s schedule.

VIII. The Trap I Am Standing In

A school that reads this essay and concludes buy the platform with the longest track record has drawn the wrong lesson, and if that school buys from me on that basis I will have sold them a false comfort.

Here is why. My longevity is a promise, backed by a record. The record is real — forty years is not a claim I have to argue for, it is a thing that either happened or did not, and it happened. But the Western Design Center is a small company, and I was born in 1945. Whatever I promise about 2045 is worth exactly as much as the institutional arrangements that outlive me, and a purchasing officer has no way to audit those from outside. A large company’s roadmap can be cancelled by a strategy meeting. A small company’s roadmap can be cancelled by a funeral. Neither of those is inspectable by a school, and a school should not have to become a credit analyst to teach a child what a sensor is.

This is exactly the problem this Foundation ran into thinking about its own perpetuity, and the answer was the same then: an arrangement that depends on the continued judgment of a living person is runtime governance. It works beautifully while the operator is sound, and it is silent about the case that actually matters.

The durable thing is not my chip. It is the fact that a student who understands what an address is, what a threshold is, and where a cycle can break has learned something no manufacturer can discontinue. If my company vanished tomorrow, that student would be fine. That is the correct design goal, and it is the one that costs me the sale.

IX. What the Foundation Will Do About It

Commitments are cheap in an essay, so these are written narrowly enough to be checked.

X. What Would Show This Wrong

That last one is written the way it is on purpose. A conflict of interest is not managed by predicting the result that costs you money. It is managed by writing down, in advance, the result that would pay you — and then letting somebody else measure.

· · ·

Coffee with Claude

The fact that did the most work here is the smallest one: that a Chromebook’s expiration is set by the platform’s launch year rather than the purchase date. It is a single sentence in a support document, it is not concealed, and it silently reprices a district’s entire refresh plan. Most of what this essay calls a sensing failure is made of facts exactly like that — published, findable, and pointed at by nobody.

One caution on my own drafting. The seven-year cadence in Section II is five data points, and five points make a pattern look more like a law than it is. The intervals could as easily reflect ordinary product-cycle economics that will not hold for the next transition. It is used here to argue that the event is foreseeable, which five points support. It should not be quoted as a prediction of 2033.

Share your understanding!