The Category Edition
Edition 33 | August 22, 2026
The gap the arguing is not closing.
Seven in ten teenagers used AI on their schoolwork this year. Three in ten have ever had a teacher talk with them about how to do it well. Common Sense Media released that survey this week, and the students turn out to be more honest about the cost than most of the adult commentary is. Thirty nine percent say they feel like they are missing out on learning when they use it. Thirty eight percent say it leaves them generating fewer of their own ideas. They can already name what it is taking from them. Nobody has taught them what to do about it.
While that gap sits there, the institutions are arguing about something else entirely. On Thursday the United States Department of Education told schools to stop judging classroom technology by what kind of thing it is and start judging it by whether it works, asking for instructional value, independent evidence, transparency for families, protection of student data, educator judgment, and outcomes you can actually measure. In the same season, procurement offices around the country are moving the other way and writing category exclusions instead, disqualifying vendors on whether a product contains artificial intelligence at all.
Both sides believe they are protecting students, and both have a real case. Neither argument does anything about the gap in that first sentence, which is the one your building will feel on Monday. Nine stories this week. Here is what happened, and what it means for your building.
The sign does not exist yet. The rule it illustrates does, and it is being written into solicitations right now.
A quiet change is moving through school procurement, and if you are not inside a bidding process you have probably not seen it yet. Agencies have begun excluding vendors by category rather than by merit. Not on price. Not on evidence of effectiveness. Not on instructional design. On whether the product contains artificial intelligence at all.
The language is specific and it is turning up in real decision letters this month. One large district professional services program has been telling applicants that it does not currently support student or teacher facing AI products, or the use of AI to process any data. That is not a line in a scoring rubric where a weaker answer costs you points. It is a threshold, and a proposal that crosses it is not read at all. The pattern is broader than one agency. Some districts now require vendors to disclose whether AI was used in preparing the proposal itself, and treat nondisclosure as grounds for removal from the solicitation. I am describing a mechanism we have been through on the vendor side this month rather than one I am predicting.
The arithmetic here is what nobody outside the industry sees. A serious response to a solicitation of this size is hundreds of hours of work: standards alignment, evidence of impact, references, insurance, privacy agreements, facilitator plans, printing and binding. Under a merit standard, a company that loses learns why and improves. Under a category standard, the entire submission is set aside before anyone reads the instructional design, and the remedy is not a better proposal. It is rebuilding the product. For an organization whose curriculum was written by humans first and had an AI layer added afterward, that is a configuration change. For the many companies that started with the model and assembled content around it, there is no version underneath to fall back to, and the rebuild is a season and a payroll.
There is a quieter casualty, and it is the one worth the most attention. When you write professional learning that will be taken by paraprofessionals, front office staff, bus drivers, cafeteria staff and teachers, you are writing for a room of adults whose needs you are not permitted to ask about. You cannot ask an adult learner what their disability is, and you should not want to. An assistant that restates a dense passage, chunks it, defines a term or slows down for whoever asks is doing that work invisibly, with no disclosure, no flag and no record. That is about as close to the spirit of universal design as software gets. Remove it under a category rule and you have not removed a feature, you have removed an accommodation layer, and you have removed it from the people for whom professional learning is already hardest. It leaves no trace, because the learners who were quietly being helped were never counted in the first place.
None of which makes the agencies wrong. A category ban is fast, legible, and defensible in a year when public anxiety is real and independent evidence is genuinely thin, and a procurement officer is not equipped to adjudicate the difference between a thoughtful application and a wrapper. It is a blunt instrument chosen for good reasons. It is still blunt.
Why it matters: Three things follow, and they are practical. If you build, find out this month whether your product can be delivered with the AI switched off, and find out before a solicitation forces the answer, because for a great many products the honest answer is no and it is far cheaper to learn that in August than inside a proposal. If you buy, understand that a category ban and an evidence standard are different instruments with different failure modes: the ban is quick and cannot see outcomes, so it will exclude the genuinely accessible option and the genuinely useless one with equal confidence. And if you lead a building, ask the question the category rule cannot ask for you, which is what happens to the people who were being quietly accommodated by the thing you just removed. Somebody on your staff was using it that way and did not tell you. That is the entire point of not having to ask.
The U.S. Department of Education released a Dear Colleague Letter on Wednesday setting out how states, districts, educators, families, and technology providers should think about classroom technology. Its central move is to separate two things that district policy usually blends: recreational device use and instructional device use are different problems and should not share one rule. From there the letter argues that screen time is the wrong primary measure of whether a tool is good or bad, and that the real question is whether the technology measurably improves teaching, learning, and student outcomes. It asks for evidence, transparency for parents, protection of student data, and educator judgment, and it leaves the decisions themselves with states and local communities rather than centralizing them. K-12 Dive read the letter as the Department stepping out in ed tech's defense at a moment when the political current runs the other way.
Why it matters: This is the most useful procurement document you will get for free this year, and it is short. The distinction it draws is the one your board keeps collapsing, because a phone in a hallway and a reading intervention on a laptop are both screens and almost nothing else about them is alike. A policy that counts minutes treats them identically and then congratulates itself. Take the letter's questions and turn them into your purchasing rubric this fall: what outcome changes, measurable by June; what independent evidence exists as distinct from a vendor case study; what families will be told; and what happens to your data if you do not renew. One caution worth holding, given the story above it. Federal guidance sets a direction, and it does not bind a state or a district procurement office, several of which are moving in the opposite direction this month.
Common Sense Media released a nationally representative survey of more than 1,000 U.S. teenagers this week, Teens in the AI Era: Schoolwork and Skills That Matter. Seventy percent report using AI for schoolwork. Among users, 63 percent say they get an answer to an assignment from AI, whether they submit it as it came (25 percent), rewrite it to sound like themselves (31 percent), or improve it with what they already know (35 percent). The students are not naive about the cost: 39 percent say they feel like they are missing out on learning when they use AI to complete assignments, and 38 percent say it leads them to generate fewer of their own ideas. The number that should stop you is the last one. Only 30 percent say a teacher has ever discussed with them how to use these tools safely.
Why it matters: Seven in ten are using it and three in ten have been taught anything about it, which means the instruction gap is more than twice the size of the usage gap, and it is the gap you can actually close. Notice also that the students are more honest about the tradeoff than most of the adult commentary is; a teenager who tells a surveyor that she is missing out on learning is describing a real internal experience and is arguably ready for a conversation about it. That conversation is cheap. It requires no license, no platform, and no procurement, and it is the single highest-return thing on this week's list. Ask your department heads a plain question at the next meeting: in which class, on which day, does a student in this building get told how to use this well. If nobody can name the class and the day, the answer is nowhere.
AI literacy became the phrase of this back-to-school season, and the Associated Press reports that schools are moving away from blanket bans toward teaching students how the technology works and where it fails. There is no settled definition of what AI literacy means or how to teach it, which is part of the story. Thirty-seven states have now published official AI guidance that districts can use as a blueprint. Meanwhile OpenAI, Google, and Anthropic all offer schools training in how to use their own tools, and a consensus is forming among researchers that this is not the same thing: good AI literacy includes knowing when not to use it. The reporting follows educators in Charleston, South Carolina, through a training session built around watching the tools make errors in front of them.
Why it matters: Hold on to the distinction buried in the middle of that story, because it will decide how well your money is spent. Training on a vendor's product teaches your staff a particular interface, and interfaces change on the vendor's schedule. Literacy teaches them to interrogate any system: where did this output come from, what is it likely to get wrong, whose data trained it, and when should a professional override it. The first expires. The second transfers. A useful test when you are handed a professional learning proposal this fall, ours included: ask whether the session would still be worth attending if the tool it references were discontinued next month. If the answer is no, you are buying product training and should price it accordingly.
A survey of the policy landscape this month finds guidance multiplying faster than anyone can absorb it, with the large majority of states having issued some form of AI direction for schools and legislators in many more introducing classroom AI bills. The pattern underneath the volume is the problem the reporting names: districts are writing rules about permitted and prohibited uses without having first articulated what they are trying to achieve, which produces documents that are precise about tools and silent about purpose. The result is a great deal of policy that will need rewriting the moment the tools shift, which they do roughly every quarter.
Why it matters: A policy that lists approved applications is a maintenance obligation you have quietly signed up for, and it will be out of date before your next board cycle. A policy that states what your district is protecting, which is usually student thinking, student data, and educator judgment, survives the product churn and gives your staff something to reason from when a situation arrives that the list did not anticipate. It arrives every week. If your AI policy is on the agenda this fall, spend the first meeting on the destination and the second on the rules, in that order, and notice how much shorter the rules get once the destination is written down.
Peninsula School District in Washington is rolling out personal AI agents for educators this school year, and Chief Information Officer Kris Hagel frames the goal in unusually plain terms: give teachers back the hours so they can spend more of them being human. The district is deliberately going top down, putting the technology in adult hands and learning from that before it writes student-facing policy. What is being deployed is agentic rather than conversational, meaning a system that pursues an assigned goal across steps and tools instead of waiting for the next prompt, which is a meaningfully different thing to supervise than a chatbot.
Why it matters: The sequencing is the transferable idea here, and it is the opposite of what most districts did in 2023. Adults first, with real support, before anything touches students, means the people who will eventually write and enforce the student policy have actually used the thing. It is also the sequence that unions and communities are most likely to accept right now. The caution worth stating clearly: an agent that acts across systems on a teacher's behalf raises a permissions question a chatbot never did, because the failure mode is no longer a bad sentence, it is an action taken in a system of record. Before you copy this, ask what your agent is allowed to touch, what it must ask permission for, and who reviews the log. Administrative time is worth reclaiming. It is worth reclaiming carefully.
FutureEd's Tara Moon profiles Washington Leadership Academy, a school far enough along the AI road to have encountered the problems the rest of the field is still theorizing about. Students understand the technology, teachers use it to do the job better, and students sit at the table when the school debates what it means. The account is notably unromantic. Not every teacher or student has bought in, and people in the building worry openly about what this means for their own job security. The most portable artifact is the school's rubric, which defines when and how students may use AI independently and what happens when they do not, pairing educational responses with disciplinary ones rather than treating every violation as cheating.
Why it matters: Two details are worth stealing. The first is the graduated response, because a student who used a chatbot to get unstuck on a hard paragraph and a student who submitted generated work wholesale have done different things, and an integrity policy with one consequence for both will lose the faculty's confidence the first time it misfires on a good kid. The second is having students in the room when the rule is written, which sounds like a nicety and is actually the enforcement strategy: rules students helped build get explained by students to each other, which is the only enforcement that scales. Note also what this school is honest about, which is that internal dissent did not disappear once the program worked. It just became discussable.
Writing for AACSB, the argument runs against the grain of most institutional AI planning: instead of asking what happens if the rollout fails, ask what happens if it succeeds exactly as designed. The piece examines the hidden cost of AI compliance in business education and lands on trust as the binding constraint, arguing for faculty AI literacy over tool mandates, for including students in the development of the policies that will govern them, and for assessments that measure whether a student can question and justify AI-generated work rather than whether they avoided it.
Why it matters: The inversion travels well below the university level. Run it on your own plan: suppose every teacher adopts the tool, every workflow speeds up, and every compliance box is checked. What did you get, and what did you quietly stop doing to get it. Most AI plans cannot answer that because they were written to manage a risk rather than to reach a destination, which is the same flaw the state policy story above identifies from the other direction. The assessment point is the concrete takeaway for K-12: an assignment that asks a student to critique and justify a machine's output is close to impossible to complete by delegating to the machine, which makes it both a better assessment and a better lesson.
UPCEA released its third annual benchmarking study of online enterprises this week, and the headline numbers look like growth: average revenues of 30.5 million dollars, up from 23.9 million last year and 18 million the year before, with 43 percent of online enterprises now returning at least two dollars for every dollar budgeted. The report's own framing is more sober than its averages. What the 2026 data shows is not a rising tide lifting every enterprise equally but a sector beginning to sort itself into institutions pulling ahead and institutions bracing for contraction, as governance centralizes and online learning stops being a side venture and becomes core academic infrastructure.
Why it matters: Averages hide sorting, and sorting is the story in every part of this field right now, including the part you work in. The same dynamic is running through K-12 professional learning: the providers with real curriculum, real evidence, and the flexibility to meet a district's constraints are consolidating, and the ones that were a thin layer over a general-purpose tool are finding out this month how thin. For a district that is a buying signal worth acting on. Ask any provider how long they have had the actual content, as distinct from how long they have had the product, and ask what happens to their offering if your board bans a category of technology next spring. The answers will sort your shortlist faster than a rubric will.
Every state-aligned iTeachAI Academy course counts toward your recertification, was written by teachers over four years, and takes about as long as a good episode of your show. AI integration, structured literacy, MTSS, SEL, and your state's mandated topics, all online, all on your schedule. Districts with restrictions can request the track that runs without any AI component at all.
One course, $25. All-access, all year, just $149.
See what counts in your state →Professional development that respects your weekend.
Two authorities gave schools opposite instructions on the same Thursday, and both of them believed they were protecting students. That is worth sitting with rather than picking a side quickly. A category ban is fast and legible and it cannot see outcomes. An evidence standard sees outcomes and it is slow, expensive, and easy to game with a vendor case study. Schools will need both, used honestly, and neither of them can do the one thing that still matters most, which is notice the learner in the room who is quietly struggling and would never say so. That was never going to be a procurement rule. It was always going to be a person.
Until next time,
Dr. Janette Camacho
CEO, iTeachAI Academy
P.S. Two free things worth doing before Monday. Read the federal Dear Colleague Letter, it is short, and lift its questions straight into whatever purchasing rubric you use this year. Then ask one question at your next leadership meeting: if we removed every AI feature from every tool in this building tomorrow, which students and which staff would lose an accommodation nobody ever wrote down. If your own recertification hours are still on the list, our catalog is at classes.iteachai.co.
Free AI courses at classes.iteachai.co
17 free AI tools at iteachai.co/TeacherTools
Know a teacher who needs this? Forward this email.
Subscribe free at iteachaibot.com
The top AI stories, tools, and practical tips for your classroom. Written by educators, for educators.
Join 1,400+ educators. Unsubscribe anytime.