The Human Ingredient Edition
Edition 34 | August 29, 2026
The tools showed up. The ingredient did not.
Khan Academy ran the experiment everyone said they wanted, and this week we got to read the results. A university researcher spent two years watching middle schoolers in a Tennessee district who had a capable AI tutor one click away, every day, at no cost. Access was nearly universal. Engagement was thin. Students opened the tutor on about a third of the days they worked in the platform, and when they did, many tried to pull answers out of it or wandered off topic. The math gains that did show up by year two, the researchers concluded, did not come from the bot.
Hold that result next to what RAND just published in its back-to-school snapshot of American education. Two-thirds of students themselves now agree that the more they lean on AI for schoolwork, the more it harms their critical thinking, and the sharpest skeptics are the students who do not use it at all. In the same five charts: teachers reporting worse well-being than comparable working adults on every measure RAND tracks, and only one in four saying they plan to stay in the profession as long as they are able. The students distrust the shortcut. The adults are running on empty. Nobody in this picture is confused about what matters.
Everywhere this week, the field kept arriving at the same missing ingredient from different directions. The researcher behind the tutoring study named human attention as the thing that makes personalized learning work. Howard Gardner argued the machines will handle more of the cognitive load than we are ready to admit, and that what cannot be delegated is how we treat each other. The HR chiefs of some of the country's largest districts drew their line at human decisions. And in Oklahoma, voters choosing who will lead their schools next advanced a candidate with forty years inside them. Ten stories this week. Here is what happened, and what it means for your building.
University of Toronto economist Philip Oreopoulos ran a two-year experiment with a Tennessee school district, tracking middle schoolers who had Khanmigo, Khan Academy's AI tutor, available inside the platform they already used for math practice. The design gets at the question district demos never answer: not whether an AI tutor can help, but whether students will actually use one when nobody is standing over them. The answer was mostly no. Students engaged with Khanmigo on roughly one third of the days they worked in Khan Academy, and the transcripts show a lot of answer-fishing and off-topic chatter rather than the patient Socratic exchange in the marketing. Students on the platform did progress faster in math than comparison groups by year two, but the researchers traced those gains to the structured practice, not to conversations with the bot. The working paper's title carries the finding: One Click Away.
The study's conclusion names the variable that the entire tutoring category has been trying to engineer around. The full benefit of personalized learning, Oreopoulos writes, depends on human attention. Motivation is not a feature you can ship. A tutor, a teacher, or a parent sitting nearby changes what a twelve-year-old does with a help button. An idle chatbot does not.
Why it matters: This is the most decision-relevant piece of evidence a district has gotten on AI tutoring, precisely because it measured behavior instead of capability. Before your next tutoring license renewal, pull the usage logs and separate three numbers your vendor reports as one: seats provisioned, students who ever opened it, and students who used it productively more than once a week. The gap between the first number and the third is what you are actually paying per engaged learner, and in this study that gap was enormous. Then notice what did work, which was structured practice with an adult in the loop. The cheapest intervention on this page is scheduling when and how the tool gets used, with a human present, rather than buying another seat of anything.
RAND has surveyed American educators for more than a decade and now runs seven standing national panels, from teachers and superintendents to students and parents. Its back-to-school brief distills the past year of that surveying into five charts, and two of them belong on every leadership agenda this fall. The first is about the adults: teachers reported worse well-being than similar working adults on every indicator RAND tracks, as they have every year since the panel began asking in 2021, with managing student behavior now the top source of stress, ahead of pay. Only one in four teachers said they plan to stay in the profession as long as they are able. The second is about AI: 67 percent of surveyed youth agreed that the more students use AI for schoolwork, the more it harms their critical thinking. Read the breakdown carefully, because it is not the pattern you would guess. Skepticism is highest among students who do not use AI for homework at all, where agreement reaches 78 percent, and older students agree most strongly of all. The remaining charts hold cautious good news: chronic absenteeism is finally falling, to a RAND-estimated 18 percent this past year, though still above the roughly 15 percent pre-pandemic norm, and restrictive phone policies appear to mitigate the distraction problem without eliminating it.
Why it matters: The AI chart quietly retires the assumption that student resistance to AI rules comes from ignorance. The students most skeptical of the technology are the ones who opted out, and the heaviest users are the least worried, which is exactly the pattern you would expect if using the shortcut dulls your sense of what it costs. Put that next to the tutoring study above: students underuse the tutor and distrust the tool, and what they are missing in both cases is not access but adult instruction and attention. Which is why the teacher chart is the one to act on first. A workforce where three in four are not committed to staying cannot absorb one more initiative that adds work. If your AI plan spends its gains on anything other than giving teachers time and attention back, this data says it will fail on arrival, and the districts in this edition that are getting it right, Fairfax and Indianapolis among them, are the ones spending AI on exactly that.
The father of multiple intelligences sat for a wide-ranging Education Week interview on what AI does to the project of schooling, and he did not offer the reassurance elder statesmen usually provide. Gardner's position is that most cognitive work schools currently certify, the disciplined mastery of subjects, the synthesis of sources, even much creative production, will be done well enough by large language models that doing it as a human becomes, in his word, optional. What he refuses to concede to the machines is the interpersonal and the ethical: how we treat other human beings, and how we act on hard questions as citizens and professionals. Schools that reorganize around what cannot be delegated, he argues, are rethinking education. Schools that do not will find the not-always-brave new world has reorganized them. His advice to working educators is plainly practical: learn the technology including its failures, experiment with it, watch colleagues experiment, and talk about what you find.
Why it matters: It is worth sitting with how much of the current curriculum falls inside what Gardner just called optional, because he is not a disruption evangelist. He has spent fifty years studying how minds work, and his conclusion converges from theory with what the Khanmigo data shows from behavior and what Fairfax's HR office says from operations: the defensible human territory is attention, judgment, and how we treat each other. For your building the actionable version is a curriculum audit question, not a philosophy seminar. For each assignment your teams give this fall, ask what the student is practicing that would still matter if a machine produced the artifact. Sometimes the honest answer is nothing, and the assignment should change. Often the answer is discussion, revision, defense of choices, collaboration, and those are the parts to grade.
Robert Franklin, an educator with more than four decades in classrooms and administration and most recently associate superintendent at Tulsa Tech, the career-training district serving high schoolers and adults, advanced in Tuesday's runoff for Oklahoma state superintendent with 55 percent of the vote. He meets Jennettie Marshall, who served eight years on the Tulsa school board, in November. Whoever wins, the office will pass to someone whose credentials were built inside Oklahoma schools, and the campaign that carried Franklin this far leaned on exactly that: four decades of showing up in buildings, not a platform of promises about them.
Why it matters: State superintendents set the weather that districts work under: standards revisions, accreditation pressure, PD priorities, and lately AI guidance, where a state chief's posture can change what your district is allowed to buy and teach. So it matters what voters said qualifies a person to hold that weather-making job, and this week's answer was time spent in schools. That is the same conclusion this whole edition keeps reaching from other directions: when the stakes are children, people trust the humans with the longest record of attending to them. For educators, it is also a quiet vote of confidence in the profession itself, and those have been rare enough lately to be worth noticing when they arrive.
Three moves in one Market Brief digest sketch where the instructional materials and professional development business is heading. CSA Education acquired BetterLesson, the teacher PD platform, through a structured asset sale, and launched BetterLesson Learning as a single organization meant to serve districts from curriculum development through classroom practice, with an AI-enabled analytics product called Abl Insights that tracks how instructional materials are actually adopted and used. EdReports, the nonprofit whose free reviews of instructional materials many adoption committees treat as a first filter, received a combined 23 million dollars from five foundations, with expanded support from Gates and Walton, continued support from Hewlett, and new money from the Lego Foundation and Valhalla. And Nucleos, an ed-tech platform serving incarcerated and reentry learners, was acquired by IT services firm iT1.
Why it matters: The through line is that curriculum and the PD attached to it are being bundled into single vendors, and the referee that scores the curriculum just got better funded. For districts the bundle is genuinely convenient and genuinely a risk: when the company that sold you the materials also sells you the training and the analytics that judge how faithfully you implemented the materials, every layer of the feedback loop has the same commercial interest. The discipline that keeps the bundle honest is independent evidence, which is exactly what the EdReports investment is betting the field still wants. Keep one layer of your adoption process, the evaluation layer, financially unrelated to the vendor being evaluated. That rule costs nothing and survives every acquisition on this page.
A panel of district human resources chiefs described how AI is entering the most sensitive office in the building, and the pattern in their answers is the story. Fairfax County's chief human resources officer, William Solomon, used AI to build a script that processes records for a workforce of more than 40,000, unglamorous work that he argues freed his team for the strategic kind. Indianapolis Public Schools built an internal tool that gives principal applicants an initial score using the same rubric human reviewers apply, with hiring committees making every final call; the district's CHRO, Christina Aden Hamer, was categorical that AI replaces no decision, and building internally rather than buying saved IPS more than 300,000 dollars this year. Colorado's education department added the warning note: without governance, AI use across an agency becomes sprawling, costly, and unproductive.
Why it matters: HR is where AI's efficiency argument meets its highest stakes, since a screening model that quietly drifts can shape who teaches in your district for a decade. What the panel modeled is the division of labor worth copying in any department: automate the moving of information, never the judging of people, and write the boundary down before the tool arrives rather than after. The IPS detail deserves particular attention from budget owners, because the 300,000 dollars was saved not by buying AI but by declining to, and building against their own rubric instead. The vendors will not tell you that is an option. It is.
With bell-to-bell phone restrictions now common, the harder question has surfaced: what about the screens the school itself hands out? District Administration surveys how systems are answering. Connecticut's state board issued guidance this month on personal technology restrictions, informed by survey data in which 57 percent of teachers called cellphone use a major classroom disruption. Florida's Hillsborough County is redefining rather than removing: its executive director of instructional technology, Sarah Garcia, describes wanting technology used sparingly and purposefully, with elementary guidance that explicitly prioritizes movement, hands-on learning, and peer interaction in grades three through five, and high school use aligned to college and career demands. The numbers driving the rethink are stark: more than 40 percent of teachers say student focus has declined over five years, and 86 percent worry about screen time's effect on mental health. Former Chicago schools chief Janice Jackson supplies the frame: avoid blanket bans, differentiate by developmental stage, judge quality of use rather than minutes, and put teachers in the room when the policy is written.
Why it matters: This is the second draft of a policy conversation the first draft got wrong in both directions, and it echoes the federal letter we covered last week: recreational screens and instructional screens are different problems needing different rules. The Hillsborough elementary language is the piece worth stealing verbatim, because it is written as a positive commitment, movement and hands-on learning first, rather than as a prohibition, which changes how teachers plan rather than merely what they avoid. The community dynamics are also worth reading correctly. Families that cheered phone bans will ask about district-issued devices next, and a leader who arrives with a developmental-stage framework before being asked will keep authorship of the policy instead of inheriting one written at a board meeting.
Veteran ESL educator Nesreen El-Baz makes an argument the AI literacy conversation has been missing: for multilingual learners, images are not decoration, they are the bridge into content, which means AI-generated visuals reach these students with extra authority and less friction. Teaching them to spot deepfake artifacts is not enough; visual access has to be paired with visual questioning. Her classroom architecture is concrete. Lateral reading, investigating a claim beyond the image rather than staring harder at it. A three-stage progression in which students interpret a source's framing and context, then generate their own AI images from different prompts to feel how prompt choices manufacture a narrative, then evaluate claims using academic language. And linguistic scaffolding throughout: students reason in their strongest language first, then present conclusions in English with sentence frames like naming what a source claims and why they questioned it.
Why it matters: The generate stage is the pedagogical move worth stealing for every classroom, not just the EL room, because a student who has manufactured a persuasive image in ninety seconds understands synthetic media in a way no detection checklist can teach. And the strongest-language-first scaffold quietly settles a policy fight many districts are having: it treats the student's home language as a reasoning asset rather than an obstacle, which is what the research has said for decades. If your AI literacy rollout this year has one unit for everyone, this piece is the argument for differentiating it, written by someone who has done it.
Inside Higher Ed rounds up five ways colleges are redesigning STEM education, from rebuilt gateway math courses, long the place where STEM ambitions go to die, to expanded hands-on learning, stronger career preparation, and new routes into technology careers. The detail that travels best is Santa Clara University's AI Kitchen, a free weekly workshop where students, faculty, staff, and Silicon Valley professionals experiment with emerging AI tools together, twelve weeks at no cost, with no grades attached. The format's premise is that fluency with new tools comes from low-stakes communal tinkering, not from a course with a syllabus and a final.
Why it matters: High schools inherit whatever colleges decide a prepared student looks like, and this list says preparation is drifting away from gatekeeping courses toward demonstrated, hands-on capability. The AI Kitchen model also scales down to K-12 almost without modification: a recurring, ungraded hour where teachers, students, and community members experiment with the same tools side by side would do more for your building's AI fluency than a mandated module, because it supplies the ingredient this whole edition keeps circling back to, humans attending to the thing together. Several of the RAND training gaps could be closed with exactly this format at close to zero cost.
Dan Fitzpatrick's back-to-school roundup catalogs what the major platforms shipped for the new year, and the pattern is that AI is moving from a separate destination into the tools teachers already use. Google's Gemini now works inside Classroom using your actual class materials as context, generating flashcards, quizzes, study guides, and shareable notebooks students can personalize, plus diagnostic-driven study plans. Microsoft's Copilot adds study coaching for students thirteen and up and unit planning for teachers. OpenAI shipped a K-12 educator plugin that builds resources and translations from approved course materials, alongside a ChatGPT for Teens with guided problem solving. Kahoot now builds activities directly inside ChatGPT, Canva's Learn Grid offers tens of thousands of curriculum-mapped resources with AI activity generation, MagicSchool added an assistant that recommends workflows from its eighty-plus tools by grade and subject, and Diffit's browser extension turns any web page into handouts and assessments.
Why it matters: Two features of this list matter more than any item on it. First, the grounding shift: the useful updates all work from your materials, your curriculum, your context, which produces meaningfully better output than the blank prompt box that burned so many teachers in 2023, and which is the difference worth checking before you adopt anything. Second, the governance implication: when AI arrives inside Classroom, Copilot, and the tools your staff already has, adoption stops being a decision anyone makes and starts being a default nobody noticed. Your acceptable-use review needs to run on this quarter's updates to existing tools, not just on new purchases, because the RAND policy gap above is about to be tested by software that installed itself.
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.
A researcher watched children ignore a tutor for two years. A survey found two thirds of students believe the shortcut is costing them something real. A theorist of intelligence spent his interview talking about respect. HR chiefs automated the paperwork and kept the judgment. And voters asked to choose a schools chief reached for the person with the most years inside schools. None of these people were in the same room, and they all found the same thing: the technology is not the scarce resource. Attention is. It always was. The systems that thrive in the next few years will not be the ones with the most tools. They will be the ones that spent their tools ruthlessly buying back human attention, and then aimed it at children.
Until next time,
Dr. Janette Camacho
CEO, iTeachAI Academy
P.S. One free exercise before Monday. Pick your most-used AI tool and pull its real usage numbers: seats paid for, staff or students who ever opened it, and people who used it well more than once last week. Those three numbers tell you more about your AI program than any vendor deck this quarter. If your own recertification hours are on this fall's 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.