We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
The week’s strongest signal is a shift in the unit of evaluation: education systems are beginning to judge AI less by model capability than by whether it preserves human learning, produces visible evidence of understanding, and fits the conditions of a real school. New York City is restricting student-facing AI and passive video access, while a Tech & Learning summary of UNESCO’s 2026 report frames AI as infrastructure that requires purpose, human agency, and governance. District evidence points to the same conclusion from the implementation side: the surrounding conditions often matter more than the tool.
Policy is becoming selective, not simply anti-technology
New York City’s policy is a concrete example. Students cannot access YouTube on school-issued Chromebooks or iPads at school or home except through teacher-shared Google Classroom lessons. The broader policy generally bars individual screen use before third grade, recommends limits of 30 minutes in grades 3–5 and 45 minutes in grades 6–8, and bans student-facing generative AI from preschool through eighth grade. Schools are being told to use paper-and-pencil tests when possible, but required digital reading, math, and state assessments remain exceptions.
The public signal is more nuanced than a wholesale retreat from technology. More than 80% of parents in one poll supported school laptops, 53% said it was important for students to learn AI tools for future careers, and 73% of registered voters still viewed excessive reliance on devices, phones, and AI as a serious problem. Parent advocates described the desired balance as guardrails, secure data, and protection from low-quality material—not zero technology. Access also remains an equity issue: support for laptops was lower and uncertainty higher among low-income parents.
UNESCO’s recommended operating model is purpose-first: identify what students or educators should be able to do better, inventory the algorithms already shaping decisions, preserve human questioning and reasoning, and redesign assessment rather than relying on detection. Its broader AI-literacy agenda includes training data, bias, hallucinations, missing perspectives, commercial incentives, privacy, consent, and human review of consequential decisions.
District implementation is emerging as the differentiator. An analysis based on 100 district-leader conversations says the districts furthest along had staged change management, capacity-building, and explicit ROI practices. Denver Public Schools moved from 225 teachers in semester-long AI professional development to 4,000 teachers in two-tier training, followed by content-area work and planned governance involving educators, families, and community members. At KIPP Public Schools Northern California, short-cycle analysis found the strongest impact from an AI instructional tool in small-group settings for students well below grade level—and no observed benefit elsewhere. The source’s conclusion is unusually specific: the same tool produced strong results in one district and inconsistent results in another, with the difference usually residing in district conditions.
The strongest product signal: preserve the learner’s work
Khan Academy is redesigning its classroom tutor around moments when learning is most at risk of becoming superficial. Sal Khan said the earlier version mostly waited for students to ask questions, even though many learners were embarrassed, could not articulate a question, or did not know what they did not understand. The newer classroom version appears after a wrong answer, asks the student to review it, and may award credit if the student demonstrates understanding. After a correct answer, it can ask for an explanation and potentially remove redundant practice. Teachers also receive narrative context—such as whether a student understood the method but made careless errors, memorized a formula, or moved unusually quickly—and Khan Academy’s planned interim assessments add prompts for students to explain their reasoning alongside conventional growth and proficiency measures.
That is a more defensible design hypothesis than “give every learner a chatbot,” but it is still a hypothesis. KIPP’s finding that benefits appeared in one setting and subgroup, rather than everywhere, reinforces the need to measure unaided transfer, subgroup effects, and implementation fidelity. The EdTech Partnerships evidence framework reaches the same standard: credible evaluation must explain whether a tool works, why, for whom, under what conditions, and with what tradeoffs for privacy, equity, affordability, interoperability, and environment.
Google’s Gemini Notebook shows a parallel move toward traceability. The product is presented as working only from uploaded files and websites, generating study guides and other materials, and attaching inline citations that open and highlight the source passage. Google says uploaded files remain private. That can make checking easier and reduce ungrounded generation, but provenance is not the same as learning impact: a source-grounded study guide still needs a teacher or learner to inspect the source and test whether understanding survives after the tool is closed.
The scarce resource is practice—and apprenticeship
The assessment problem is becoming a learning-design problem. UNESCO describes a “classroom AI doom loop” in which AI generates the assignment, students use AI to complete it, AI evaluates the response, and an automated system records the grade. Its proposed alternative is to make thinking visible through oral explanation, iterative drafts, authentic projects, reflection, in-class problem solving, and documentation of AI use.
Ethan Mollick makes the same issue concrete for both education and work. He argues that learners can outsource repetitive practice because the work feels difficult, losing the basic skill they were supposed to build. He also warns that the entry-level apprenticeship pipeline can break when interns use AI to perform work they are still learning, managers route work around them, and companies prize efficiency over the inefficient periods in which novices receive human feedback. His proposed correction is to make teaching central and retain some assessments and work periods completed without AI.
Justin Reich adds an important caution against premature certainty. He argues that educators know far less about effective practice with large language models than they do about established areas such as reading instruction, so the near-term response should be systematic experimentation rather than confident “best practices.” In his view, output evaluation depends heavily on domain knowledge, because users need to recognize when a model’s fluent answer is wrong. He also says school change spreads primarily through peer learning, not by distributing software or issuing a policy.
AI literacy and higher education are moving toward guided practice
Experience AI, the free Raspberry Pi Foundation–Google DeepMind curriculum, offers a practical response that does not require teachers to become AI specialists. Its approach embeds AI literacy in existing work with data, prediction, context, research verification, critical thinking, and problem solving. One age-banded activity has 8- to 12-year-olds analyze misleading advertisements that anthropomorphize AI; the curriculum stresses that systems make predictions, are not people, and leave humans accountable. The program says it has been downloaded more than one million times, supports 22 languages, works with more than 40 international partners, and is being used by some schools in weekly seventh-grade advisory.
Higher education is confronting the same shift from content delivery to experience, judgment, and adaptability. Michael Horn argues that college may become more valuable when AI can teach information anywhere, because live interaction, research, relationships, experiential learning, networks, and knowing when knowledge applies become more important. A separate higher-ed analysis argues that policies, task forces, and isolated pilots do not create agility; institutions need the ability to move faculty, technology, curricula, pedagogy, assessment, and governance quickly.
There are early institutional versions of this model. Northeastern is scaling faculty cohorts from roughly 20 participants to an estimated 100–200 across colleges, tying AI work to co-op and disciplinary conversations and distinguishing AI literacy from AI fluency. Its faculty-development discussion rejects formal AI detection because of privacy and false-accusation risks, emphasizing clear expectations and redesigned pedagogy instead. Khan described a proposed KNED Institute in similar terms: authentic, time-constrained work, global groups, peer review, AI-project portfolios, and learner-selected evidence for employers. The interest list exceeds 30,000, but that is demand for the model—not evidence that it works.
A global access example makes the same point without generative AI. In South Africa, Worldreader’s data-free BookSmart evaluation combined a four-group randomized trial with caregiver interviews: 812 caregivers enrolled and 519 completed both baseline and endline assessments. Completing more books was associated with stronger caregiver-reported expressive-language skills, while goal-setting messages increased completion of accompanying activities more than reminders. Engagement remained modest, families requested more home-language books and clearer routines, and the preliminary findings have not been peer reviewed.
What This Means
- For edtech buyers: Treat every AI product as a learning hypothesis. Define the target capability, baseline, subgroup, implementation conditions, transfer task, and post-adoption review before counting usage as success. The relevant evidence standard is efficacy plus effectiveness, ethics, equity, and environmental viability.
- For assessment leaders: Separate practice from certification. Let AI support formative feedback, critique, or scaffolding where appropriate, but require students to explain, revise, solve unfamiliar problems, or demonstrate skills without the system when the goal is to certify competence. Do not make an imperfect detector the primary judge of authorship.
- For K–12 leaders: Replace blanket “AI” categories with age-, task-, and tool-specific rules. A ban on generic student-facing AI can coexist with narrow, purpose-built tools, but only if schools define the exception, protect access and privacy, and plan for the paper, device, accessibility, and assessment consequences of the transition.
- For teachers and L&D teams: Teach discernment inside subject learning. Ask whether AI is helping learners notice, question, practice, and defend their thinking—or merely removing the difficulty that builds the skill. Experience AI’s subject-embedded model and Mollick’s apprenticeship warning point in the same direction.
- For higher education: Build agility and applied experience as infrastructure, not as a one-off AI committee. Co-op, authentic projects, peer feedback, and portfolios can make learning visible, but they need outcome evidence and employer validation rather than relying on novelty.
Watch This Space
- Khan Academy’s classroom tutor and interim assessments: The important next evidence is unaided performance, durable mastery, and subgroup results—not simply whether students engage with the tutor or complete more generated practice.
- New York City’s implementation: Watch how schools reconcile paper-first guidance with required digital assessments, how exceptions are governed, and whether restrictions are accompanied by usable AI literacy rather than access control alone.
- District evidence infrastructure: The market is moving toward teacher coaching and institutional experimentation—McGraw Hill acquired TeachFX, while Cornell is launching faculty-led AI pilots and student fellows—but the meaningful test is whether these efforts produce comparable evidence across schools and learners.
- Higher-ed and workforce signals: Follow whether portfolio-based, AI-enabled programs and co-op-linked faculty work become credible hiring signals while institutions preserve the human feedback and deliberate practice that entry-level workers need.
- Global digital learning: BookSmart’s peer-reviewed follow-up should clarify whether data-free access, local-language content, and goal-setting can turn availability into sustained family reading.
- Khan Academy’s AI tutor is shifting from optional help to proactive classroom support. Sal Khan says the earlier version mostly waited for students to ask questions, while a new classroom version launched about a month before the interview after a prior pilot and now intervenes after wrong answers to prompt review and explanation. The planned next step is to credit students who demonstrate understanding or reduce redundant practice, while giving teachers context to distinguish understanding from careless errors, memorization, or possible cheating.
- Khan Academy is expanding assessment beyond percentile and growth scores. Its interim assessments are slated to enter the market in the fall and are designed to be psychometrically valid, transparent, accessible, and actionable. Students can be asked to explain their reasoning, with reports adding narrative detail about understanding, problem-solving approach, and careless errors so educators can adjust classroom rotations or group work.
- Schoolhouse’s volunteer-tutor pipeline combines human verification with AI practice and feedback. Candidates record their face and screen while explaining their reasoning, must meet a 90% threshold, and undergo peer review before certification. After certification, tutors practice with AI simulations, receive feedback, are observed by senior tutors in early live sessions, and get ratings from learners; the AI coaching is kept separate from the public tutor rating.
- The planned KNED Institute is designed around practice-based, portfolio-oriented online education. Khan says it has publicly announced a bachelor’s program, attracted an interest list of more than 30,000 people, and may add certificate and master’s programs, with the certificate potentially launching within six months and the master’s within a year. The proposed model uses globally distributed video groups, authentic work tasks, peer review, and portfolios showing AI projects and durable skills, while allowing learners to choose which evidence to showcase.
- Khan identifies cognitive offloading as a central AI-in-learning risk. He says people are already using AI for cheating and life decisions, but also describes AI and online guidance giving him the confidence to solve unfamiliar practical problems himself; in his view, AI can either weaken or strengthen independent capability depending on how learners use it.
- Alpha School reports moving academic instruction from classroom teachers to a personalized learning platform that assesses students and creates individualized lesson plans, allowing educators to focus more on coaching and life skills such as leadership, teamwork, relationship-building, financial literacy, and public speaking. Alpha claims students arriving around the 30th percentile and those at the 99th percentile can achieve growth rates in the 80s and 90s. It uses NWEA MAP assessments three times annually and said MIT and Harvard were preparing independent research and evaluation; those results were not reported in the discussion.
- Alpha’s self-directed “brain lift” model gives high-school students 45 minutes daily to pursue a research question of their choice. One example involved a student researching teen dating, building an LLM to provide advice, and comparing its output with ChatGPT and 150 mothers; the speaker said the resulting study had been submitted to Nature and was in final peer review.
- Panelists distinguished narrow, purpose-built AI (“teach, not tell”) from general-purpose or “wide” AI, citing handwriting-problem detection, new assessment methods, and support for learning differences as promising applications. They warned against giving young people general-purpose chatbots or AI companions that are not designed for learning or child safety. A Florida K-12 AI task-force participant described districts combining privacy and child-protection measures with performance-based contracts for purpose-built tools. Utah’s state superintendent argued that both blanket bans and blanket adoption are inappropriate, urging decisions based on measured learning outcomes and implementation at the school and teacher level.
- AI is driving a shift from essay-only assignments toward knowledge application and demonstrated mastery: speakers advocated preserving foundational content knowledge while adding applied work, and Alpha described requiring students to give a five-minute talk and answer questions alongside a written paper. Panelists said teacher preparation and pedagogy will also need to change. Recommended AI literacy includes identifying hallucinations, bias, and the difference between fact and opinion; one panelist estimated that about 80% of this instruction could be done without screens. Another argued that upper-secondary students should gain direct experience with AI tools, while younger learners should face a higher evidence bar before deployment.
- Justin Reich argues that schools should treat LLM use as an unsettled practice: educators know far less about what to do with LLMs than with established areas such as reading instruction, so they should acknowledge uncertainty, run systematic experiments, and evaluate practice as it evolves. He says many AI-literacy frameworks, teaching strategies, and prompting strategies currently presented as answers lack good evidence, creating a risk of converging on ineffective practice.
- Effective LLM use may depend less on prompting than on evaluating outputs: models produce uneven, non-deterministic responses, and domain knowledge helps users distinguish useful work from bad or incorrect work. Reich asked engineers at DeepMind how to teach a junior engineer to code with a copilot, but none could answer, illustrating that AI-enabled competencies are not yet sufficiently codified for straightforward classroom transfer.
- Technology distribution alone rarely changes pedagogy: teachers initially use new tools to extend existing practices, while meaningful change requires years of experimentation plus changes to curriculum, routines, IT environments, and family support. Reich also warns that new technologies tend to benefit affluent communities more, where learners receive more interesting tasks and adult support. At school-system scale, execution matters more than choosing a supposedly perfect approach, and teachers most often change practice through peer learning.
- Reich's project based on 120 interviews with teachers and students found that ChatGPT's educational importance varies sharply by context: in wealthier communities it was often a major source of problems and opportunities, while in lower-resource communities it ranked behind urgent issues such as hunger, homelessness, and students not attending school—described as 19th on the list of problems.
- The discussion distinguishes AI substitution from augmentation: when AI takes over a task, people may lose practice in that skill through “task offloading,” whereas using AI as a sparring partner or scaffold is presented as a more actively engaged mode.
- It argues that the studies under discussion mostly measure substitution rather than augmentation, while measuring cognitive gains from collaboration with AI is harder because it requires longitudinal tracking and measures beyond IQ, such as synthesis and meta-reasoning.
- The practical learning recommendation is to use AI to make learners articulate, challenge, refine, and defend their thinking—not make decisions for them—summarized as asking AI to “think with you” rather than “think for you.”
- Demis Hassabis said AI’s rapid development is already making three-year computer-science course design difficult because the underlying material can change in less than three years. He recommended maintaining strong coding and mathematics foundations, actively experimenting with AI tools, and developing the ability to learn new information quickly; he expects natural-language programming to broaden access to coding while amplifying engineers’ output.
- Hassabis urged universities to pursue multidisciplinary work that applies AI to specialist fields, arguing that substantial advances could come from this approach over the next five to ten years.
- AlphaZero offers a concrete example of AI feeding back into human skill development: after discovering new chess strategies and styles, its games and associated books influenced world champion Magnus Carlsen, who incorporated AlphaZero’s ideas into his own play.
- AI can undermine foundational learning when students outsource repetitive practice because learning feels difficult; Mollick argues that schools and companies must help learners decide which skills to retain and which work to delegate.
- AI tutoring is already powerful, but education is still working out the right balance between AI-supported tutoring and assessments completed in classrooms. Mollick’s practical boundary is that people should not delegate work to AI when their goal is to learn—unless they are using the AI actively as part of the learning process.
- AI is disrupting the entry-level apprenticeship pipeline: interns can use AI to perform work they are still learning, managers can route work around them, and companies may prioritize efficiency over training. Mollick says organizations need to deliberately preserve inefficient learning periods, including human feedback and assessments completed without AI; universities may also need to provide additional workplace-oriented training.
- Andrew Ng said fear-driven messaging is already affecting students: he sees high-school students considering not using AI because they fear it could cause human extinction, a concern he says could slow AI development.
- Rapid AI-driven labor-market change is increasing the need for continual reskilling: the segment reported about twice as many tech layoffs this year; coding roles once attainable by learning a few languages are being automated, while new roles require AI-related skills and workers struggle to catch up as models and skill requirements change daily. Younger workers were described as especially vulnerable because their skills are harder to match to the new labor market.
- Andrew Ng called the AI industry’s fear-heavy messaging “very problematic” and said he sees high-school students considering avoiding AI because they fear it could lead to human extinction.
- The segment describes an AI-driven reskilling challenge: workers are struggling to understand how their existing skills fit the new labor market, coding work is described as increasingly automated, AI-linked roles require unfamiliar skills, and workers say a new model or skill appears every day, making it difficult to keep up. Younger workers are described as especially vulnerable because their skills are harder to match to the changing market.
- Demis Hassabis said the pace of AI development is making even three-year computer-science courses difficult to design because the underlying material changes in less than three years. He recommends retaining coding and mathematics fundamentals to use AI tools more deeply, while natural-language programming could broaden access to creating games, films, and applications and amplify engineers’ output.
- Hassabis advised learners to experiment with AI tools in their spare time and train themselves to acquire new information rapidly.
- He urged universities to pursue multidisciplinary work that applies AI to specialist fields, anticipating substantial advances from this approach over the next five to ten years.
- Alpha School uses GPT as a coaching and quality-assurance layer for its weekly one-to-one “limitless meetings”: guide and student wear lanyard recorders, conversations are transcribed, and GPT analyzes speaking time, question-versus-statement balance, and who is driving the discussion; Alpha says the recordings remain within its network.
- The AI augments rather than replaces the human educator: students continue speaking with a guide, while GPT supplies higher-order prompts and feedback intended to make conversations less shallow and more student-led. One guide said his first two meetings were rated “below expectations” because he was leading too much and asking overly directive questions, giving him concrete feedback to increase student autonomy.
- Alpha also uses the recordings and GPT analysis for oversight of one-to-one interactions, alongside a rule that meetings occur in visible, non-closed-door spaces; the school says its tools check for inappropriate or unsafe interactions. The underlying practice is a 10–30-minute, goal-setting conversation used from pre-K through high school, with the intended progression toward students bringing their own agendas and leading the discussion.
- Sam Altman described a rapid progression in AI mathematical capability: from barely handling grade-school word problems three summers earlier, to performing well at AIME the next summer, winning a gold medal at the IMO one summer earlier, and proving one of the seven biggest unsolved problems in mathematics by the current summer.
- Salesforce paired its Claude plugin rollout with hands-on AI enablement: the plugin is in open beta, remote users are directed to Trailhead, and Dreamforce attendees can use AI learning labs to learn how to build a personalized command center and leave with their organization enabled.
- Salesforce’s experience frames agent building as a new workforce skill because probabilistic systems are difficult to build with; to reduce that barrier, it is offering out-of-the-box agents that business users—not only IT teams—can implement and deploy.
- Daphne Koller says Coursera achieved scale and democratized access, but its early use of prerecorded video snippets and quizzes did not deliver the personalized teaching associated with the “Bloom two sigma” performance effect; she hopes current AI, paired with education-specific data collection, can reach that level of personalization.
- If launching Coursera today, Koller says she would probably start with an AI tutor rather than a course platform. She does not consider lectures obsolete, because exceptional teachers can inspire and motivate learners, suggesting a future that combines AI tutoring with human-led instruction.
- Koller argues that universities should make their boundaries more porous, enabling greater movement of people, ideas, and intellectual property between academia and industry so research can have more societal impact; she says institutions are still working out how to balance faculty mobility, entrepreneurship, and academic roles.
Andrew Ng reports that fear-focused AI messaging is affecting student attitudes: he has seen high school students consider avoiding AI because they believe it could lead to human extinction, and argues that this negative sentiment could slow AI development.
Alpha School founder MacKenzie Price argues that education should shift from parents manufacturing a predetermined outcome to helping children become who they are: her admissions-interview experience contrasted students driven by parental expectations with students motivated by their own curiosity, leading her to favor nurturing, coaching, support, and high standards without controlling the destination. In the school model presented as disrupting education through AI, Price says family alignment matters more for younger students, while high-school students’ own preferences should carry more weight; Alpha’s stated philosophy combines high standards with high support and lets each child find what is right for them.
The conversation also highlights a practical shift away from overparenting: children may lack emotional-regulation and everyday life skills when adults solve too many problems for them, while Alpha’s parent guidance is to “support, not solve,” helping learners develop independence through listening, advocacy, and experience.
- Salesforce and Anthropic opened Claude Force in beta and offered Trailhead materials plus hands-on AI learning labs where users could learn to build and activate their own AI command centers.
- Salesforce demonstrated Agentforce’s Marshall learning supplier-onboarding workflows in an SAP sandbox described as a “classroom.” The agent used LLM reasoning to learn required fields and business rules in 90 minutes, versus weeks for a new employee, then converted those learnings into reusable trusted actions. In the demonstration, this enabled autonomous SAP updates and reduced supplier onboarding from days to hours.
- District readiness is the main determinant of AI implementation: Across more than 100 district-leader conversations, AI had a focused role in day-to-day work where districts already had strong change-management, capacity-building, and ROI systems. The same AI tool produced strong results in one district and inconsistent results in another; the author says the difference was usually the surrounding district conditions rather than the tool itself.
- Districts are moving from one-off AI rollouts to structured professional learning and governance: Denver Public Schools progressed from 225 teachers in semester-long AI professional development to 4,000 teachers completing AI 101/201, followed by job-embedded content-area work and planned governance involving educators, leaders, families, and community members. Bullitt County Public Schools built custom chatbots aligned with its instructional framework, had a teacher coalition test them, and was establishing lab classrooms to make implementation observable.
- KIPP Public Schools Northern California found a targeted, data-backed learning effect: Using a privacy-preserving application built with Claude Code, the team compared interim assessments and student demographics with four months of usage data from an AI instructional tool. The strongest impact appeared in small-group settings for students well below grade level, with no observed benefit in other settings.
- AI education products and institutional adoption are expanding: McGraw Hill acquired TeachFX, an AI instructional-coaching platform that analyzes classroom audio and gives teachers feedback on questioning and student discussion while providing aggregated insights for coaching and curriculum implementation.
- Higher education is pairing experimentation with calls for stronger safeguards: Student Defense reported that colleges are integrating AI across the student experience faster than they are establishing consistent protections, and called for transparency, human oversight, data protections, and institutional guardrails. Cornell is launching faculty-led AI Integration Pilots and expanding its AI Undergraduate Fellows program so students and faculty can evaluate where AI should be used and its effects on learning and classroom experience.
- Salesforce and Siemens demonstrated an AI agent learning supplier-onboarding workflows in an SAP sandbox described as a “classroom”: Marshall used LLM reasoning to learn required fields and business rules, including country-specific documentation, in 90 minutes—compared with weeks for a new employee—and converted the result into reusable “trusted actions.” The same system was presented as reducing supplier onboarding from days to hours through autonomous execution with human review when needed.
- An open-ended Claude task asked the model to investigate the Voynich Manuscript and create a shareable movie. Claude failed to solve the mystery but produced a movie described as an interesting, useful explainer—an example of AI-generated explanatory media for self-directed learning, with no evidence here of learner outcomes.
- The experiment also exposed important limits: the model tried 211 techniques, including 31 it claimed were novel, but all reached dead ends; a guardrail triggered a switch to Opus, whose resulting writing was described as “too Claudy.” The full multi-agent analysis and movie would have cost about $85 in Fable 5.1 outside the stated usage limit, and the experiment cautioned against anthropomorphizing the model.
Khan Academy is promoting free, four-week study plans for Texas students covering math, science, and financial literacy courses.
