Labor Matters: AI, a Natural-Born Teacher
Hundreds of Organizations Are Using AI to Teach. The Burning Glass Institute AI Use Case Dataset Shows Who, and How.
Opening
AI is not just changing how people work. It is beginning to change how people learn.
The most obvious place to look for that change is education technology: tutoring apps, language-learning platforms, online universities, test-prep companies. But that turns out to be too narrow. AI is also being used to train sales representatives, coach job candidates, give feedback to call-center workers, help nurses practice clinical judgment, teach music students, support student drivers, prepare athletes, and personalize instruction for students with disabilities.
In other words, AI as teacher is not confined to schools. It is spreading across the economy.
That is the surprise this report is built around — and the reason the Burning Glass Institute AI Use Case dataset is useful. Most ways of measuring AI adoption are indirect. Some estimate where AI could be used by mapping AI capabilities to tasks and occupations. Others ask executives whether their firms “use AI,” count customers of AI vendors, or project future spending. These approaches are valuable, but they mostly capture theoretical exposure, self-reported adoption, sales activity, or expectations.
AIUC is different. It measures observed adoption: instances where organizations describe AI being put to work in job postings, career-history profiles, and other public sources. It does not ask where AI might matter in principle. It records where AI appears in the actual documentary trail of work.
Online job ads tell you more than what a company wants in a hire. Most describe the work itself: the project the candidate will join, the systems being built, the AI being deployed. Career-history profiles work the same way in reverse, narrating what a person actually did: the model they fine-tuned, the recommendation engine they shipped, the chatbot they stood up, the simulator they built.
None of these documents was written to describe an AI market. That is precisely why they are valuable. They record AI as a fact of the work, not as a marketing claim.
AIUC turns those traces into structured data. It extracts mentions of AI deployments from this corpus and normalizes each into a standardized sentence describing a single use of AI: who is deploying it, what it does, and what it acts on. Around that sentence sits a layer of tags — the action the AI performs, the domain it operates on, the techniques behind it, the occupation and industry it sits in, the source type and date, the location, and the specificity of the underlying evidence.
The result is a structured, growing record of AI moving into the actual work of the economy. This report asks a single question: who is using AI to directly teach, tutor, coach, or give feedback to a human learner?
The answer is a remarkable range of organizations, far beyond the edtech companies you would expect. The catalogue covers ten categories: K-12 tutoring, corporate workforce training, online higher education, university teaching, language learning, test preparation, interview and speaking preparation, special education, sports coaching, and a residual “other” bucket covering music students, student drivers, and homeschoolers. It surfaces hundreds of organizations and products from text that never set out to describe an education market — which is exactly what makes the breadth credible.
The catalogue focuses on AI in a learner-facing instructional role. It excludes teacher-productivity AI such as lesson planning and grading, institutional analytics such as admissions or retention prediction, pure operations, tutor-matching marketplaces, and AI that supports recruiters in screening candidates.
Part I — The Buckets
1. K-12 Tutoring
The learner is a K-12 student — kindergarten through twelfth grade, at a classroom workstation, at home, or in an after-school program.
AI walks the student through a problem and refuses to give the answer, offering progressive hints scoped to the smallest next step. It re-serves the flashcards they keep getting wrong, varies question type by inferred mastery, and auto-generates flashcards and practice tests from pasted notes, slides, or PDFs.
On math, it watches the strategy the student uses to reach an answer — clicks, keystrokes, time on task — and evaluates the strategy itself, not just whether the final answer is right.
On reading, a speech-recognition model trained on children’s voices listens to the student read aloud, flags the words they stumble on, and picks from a library of evidence-based micro-interventions when they stall.
An eye-tracking screen identifies risk of dyslexia in under five minutes. When the student works with a human tutor, AI matches them to the right one and augments the live session with transcription, summaries, and follow-up practice.
The organizations doing this include established edtech vendors layering generative AI on top of existing adaptive engines, large virtual-school operators with AI tutors and AI teacher assistants, tutoring marketplaces using AI to scale a labor-intensive service, and university research labs whose decades of intelligent-tutoring-systems work feed the commercial products.
2. Corporate Learning and Development
The buyer is the employer. The learner is the employee — a software engineer being upskilled in generative AI, a sales rep practicing a discovery call, a nurse working through compliance training, a new hire onboarding into a Fortune 500 role.
AI helps the employee in two main ways.
First, a personalized learning-path engine ingests their job role, skills, project history, and performance signals, then orchestrates a sequence of microlearning, courses, certifications, and stretch assignments. Some platforms layer in human mentor matching alongside automated assessment.
Second, a simulation engine puts the employee into an LLM-generated role-play — most often a customer or prospect conversation, sometimes built from sales decks they upload — captures the response with speech-to-text, and grades both substance and form: pace, filler words, objection handling, soft skills, next-step discipline. In many cases, the system returns a scorecard almost immediately.
The supply side is dominated by HR-tech vendors, consultancies, and content platforms selling employee-learning AI to other employers. The largest publicly disclosed AI-skills training commitments come from this group: billion-dollar vendor commitments, large-scale platform investments, and major technology-company pledges to train millions of learners in AI. Sales-coaching specialists, learning-experience platform operators, and high-stakes professional-training providers — including aviation simulator training and legal continuing education — round out the supply side.
The demand side is even more important. Large enterprises deploying these tools in-house, often building their own AI overlays, are by a wide margin the largest cohort of learner-facing AI buyers in the corpus. They span healthcare, financial services, technology, retail, energy, hospitality, telecommunications, and consumer goods. Many Fortune 500 employers appear in the broader corpus.
This is the most important finding in the report. The earliest large-scale institutional buyer of learner-facing AI is the employer.
3. Online Higher Education
The learner is a university student taking some or all of their coursework online — a self-paced MOOC learner, a competency-based-degree student, an undergraduate in a gateway biology lab augmented by VR, or an online master’s student getting answers from an AI teaching assistant on a course discussion forum.
AI gives the student a personal tutor that answers only from the specific lecture video, reading, or assignment they are looking at. It gives them an organization-wide generative-AI deployment, embedded into coursework, advising, and self-directed study. In some cases, it pairs adaptive courseware with VR-based science labs. And it gives them a conversational AI teaching assistant in online graduate courses that answers questions at any hour and routes harder cases to a human.
The organizations include major MOOC platforms, large online and competency-based universities, and institutions deploying generative AI at organizational scale.
4. University Teaching
The learner is a residential university or community-college student in a face-to-face or hybrid course.
AI helps the student in two main ways. An adaptive-courseware product embedded in their textbook serves personalized practice and reading, reports progress to the instructor, and adapts what they see next. A direct AI tutor gives them 24/7 explanations and step-by-step solutions outside of class — sometimes branded around a prominent instructor, sometimes as a general homework-help service spanning math, science, business, humanities, and writing.
The organizations include textbook publishers and institutional adopters of adaptive courseware, AI-product companies built specifically for the residential teaching mission, online “webtext” providers serving hundreds of U.S. universities, and the commercial heirs of the original academic intelligent-tutoring-systems lineage.
5. Language Learning
The mass-market learner is someone trying to acquire or improve a non-native language, most often English — a Korean teenager preparing for U.S. universities, a Brazilian engineer working on accent, a French executive practicing conversation.
AI scores pronunciation against a reference phoneme by phoneme and tells the learner which sounds to drill next, with side-by-side playback against an AI reference voice. It lets them practice unscripted conversation — small talk, ordering food, debating a topic — without a human partner. For accent reduction, it analyzes a baseline assessment of the learner’s speech against the target accent’s sounds and builds a personalized daily-lesson path.
Three groups of organizations build this. Consumer pronunciation and conversation apps dominate by volume. A second group operates AI-administered language-proficiency exams, combining adaptive testing, AI scoring, Item Response Theory weighting, and human-in-the-loop review. A third layer uses similar technology for defense, intelligence, and military-linguist training in languages such as Levantine, Yemeni, and Iraqi Arabic.
6. Test Preparation
The exam is high-stakes. The learner is preparing for a standardized academic test, a graduate or professional licensure exam — medicine, nursing, law, accounting, finance — or a corporate technical certification in areas such as cloud platforms or customer-relationship-management systems.
AI helps the test-taker in two specific ways. An adaptive question bank serves practice items matched to current weak areas and generates a written explanation of why each missed answer was wrong. An automated scoring engine grades the essay or spoken response against rubrics derived from human raters — sometimes as the only grader, sometimes as a check on a human grader, with disagreements escalated for review.
The organizations include major testing companies that operate the exams themselves, dominant question-bank operators in graduate professional exam preparation, platform vendors generating certification practice content for cloud-vendor and CRM-vendor ecosystems, and consumer test-prep brands.
7. Interview and Speaking Preparation
The workflow is tight. The learner is a job candidate preparing for an interview, or a professional preparing to give a presentation or sales pitch.
The candidate records a practice answer or rehearses a talk. The system transcribes the speech with speech-to-text, scores it against a rubric — relevance, structure, filler words, pace, eye contact, body language, sensitive language, monotone pitch, originality, inclusive-language flags — and returns a written feedback report. For mock interviews, an LLM plays the interviewer in character. For sales coaching, the LLM plays the prospect. Body-language scoring uses webcam-based head-pose detection in at least one widely deployed productivity-software product. Some products use AI digital-human avatars as the coaching surface.
The organizations include AI interview-prep startups, large productivity-software vendors embedding speaker-coaching into existing products, outplacement firms and large employers that bundle the tools into their offerings, and university career-services functions. Many products are thin LLM layers on top of speech-to-text infrastructure; the differentiation lives in rubric design, role-specific prompts, and distribution.
8. Special Education
The learner is a student with a specific learning difference, disability, or developmental condition — most commonly dyslexia, autism, ADHD, or a speech impediment.
AI helps the student in ways tightly matched to the condition. A children’s-speech-recognition model listens to the student read aloud and flags the exact word-level errors a dyslexia clinician would flag, then picks the next intervention from an evidence-based library. A text-to-speech reader reads any document aloud while highlighting words in sync, for learners who decode more easily by ear than by eye. A face- and emotion-detection module prompts a child with autism to recognize and respond to social cues. A humanoid robot delivers structured social-skills interactions in autism therapy. A speech-recognition algorithm gives a child with a speech impediment immediate articulation feedback.
The organizations include small specialist vendors with clinical evidence bases and university research labs working in human-robot interaction, speech therapy, and adaptive therapeutic systems for special education.
9. Sports Coaching
The data substrate is movement video. The learner is an athlete — a recreational golfer working on a swing, a high school football team reviewing game film, a tennis player tracking serves, a Pilates student getting form feedback, or a pitcher refining mechanics.
AI helps the athlete in two specific ways. A phone-based pose-estimation or motion-capture model converts video of movement into quantitative metrics — hip rotation, shoulder turn, swing tempo, stroke depth, joint angles — and shows where the athlete deviates from a target. A team-level video-analysis platform automates the tagging and highlight generation that used to take a coach hours of film review, then feeds the output into player-development workflows across football, basketball, soccer, volleyball, and hockey. Tennis-specific products use phone or smartwatch sensors for automatic scoring, shot tracking, line calling, and stroke analytics.
The organizations include sports-specific AI startups, incumbent equipment makers adding AI features, professional teams and university athletic departments deploying these tools internally, and cross-discipline players in baseball player development and movement-quality measurement.
10. Other AI Teaching
Three smaller learner populations round out the catalogue: music students, student drivers, and homeschoolers. Each is thin in the corpus — research universities and a single specialist vendor in music; two online driver-ed vendors; a small set of homeschooling and micro-school platforms. The thinness is itself a signal. AI music coaching, AI driver-ed, and AI homeschool platforms exist at the research-and-early-product stage, not as mature categories at the scale of sports coaching or language learning. The long tail is real, but it is still mostly a tail.
Part II — Patterns
The catalogue is broad, but it is not random. Four patterns explain where learner-facing AI is showing up first.
1. The buyer is often the employer, not the school.
The largest single cohort of learner-facing AI deployment in the catalogue is not K-12 districts or universities. It is employers: sales reps practicing discovery calls, software engineers being upskilled in generative AI, nurses working through compliance modules, new hires onboarding into large organizations, and managers receiving personalized learning paths.
This changes how we should think about AI in education. The public conversation often imagines AI teaching as a school story. That story is real, but incomplete. Employers have large training budgets, measurable performance outcomes, and a constant need to reskill workers. AI as teacher is not only an education-sector phenomenon. It is part of the infrastructure of workforce development.
2. AI teaching advances fastest where performance can be measured cleanly.
The densest buckets are the ones where AI has something concrete to evaluate. Pronunciation can be scored against phoneme alignment. Athletic motion can be compared against pose-estimation benchmarks. Math problem-solving can be evaluated through strategy traces. Oral reading can be assessed through speech recognition. Test responses can be graded against rubrics. Sales conversations can be scored for pacing, structure, objection handling, and next-step discipline.
Learner-facing AI is not spreading evenly across all forms of learning. It is spreading first where learning can be broken into observable performances, repeated practice tasks, and measurable feedback loops.
3. The instructional shape is practice, feedback, and simulation — not lecture replacement.
Across the catalogue, AI rarely appears as the main lecturer. It more often appears as the patient drill partner, rubric grader, role-play counterparty, pronunciation coach, mock interviewer, speech evaluator, or tutor that gives the next hint.
That distinction matters. The most important near-term role for AI in learning may not be replacing teachers or professors. It may be filling the gap between exposition and mastery: the repeated attempt, the immediate correction, the targeted drill, and the realistic simulation. The emerging pattern is not AI as lecturer. It is AI as coach.
The coach role is the entry point. The teacher role is next.
4. AI teaching is a function, not a single product category.
The strongest examples are not usually standalone chatbots dropped into learning environments. They are AI tools embedded into existing systems: corporate learning platforms, sales enablement tools, test-prep question banks, language-learning apps, online course platforms, adaptive textbooks, sports video systems, career-services tools, and special-education interventions.
That is what AIUC makes visible. Individually, each example can look anecdotal: an AI tutor here, a sales simulator there, a pronunciation coach somewhere else. Taken together, they show a larger pattern. AI is entering learning through the feedback loop — and that loop can be inserted into many different institutional settings.
Closing
AI is not waiting for the education sector to invite it in. The first big institutional buyer is the employer. The first wide deployments are in performance-measurable domains. The first instructional shape is the coach, not the lecturer. The school version of this story is coming. The economy-wide version is already here.