I've written previously about the technical definition of adaptive learning and about why most platforms don't meet it. This post applies that framework to the three biggest brands in the space — Pearson, Coursera, and Khan Academy — and looks at what each actually ships in production. The conclusion isn't subtle. None of them clears the technical bar. The reasons are different in each case, and they're worth understanding because they predict where the category goes next.
A note before I start: I have no inside information on any of these companies' internal roadmaps. Everything here is based on public statements, published product behavior, and what's observable from being a paying customer. If any of these vendors have a real adaptive engine that I'm missing because they haven't shipped it yet, I'd genuinely love to be corrected. But I've looked, and I don't see it.
Pearson
Pearson is the publishing giant of the legacy edtech world. Their revenue is roughly $3.7B against a market cap that's been bouncing around $5-6B for years. They've been telling investors a story about an "AI-powered personalized learning" strategy for at least three years now. The most prominent execution of that story is Pearson+, a subscription bundle of their textbooks and assessments, marketed as an adaptive learning platform.
What Pearson+ actually delivers is digital access to traditional textbook content, with a quiz layer that uses item-response-theory difficulty estimation. There's a "personalization" feature that suggests practice problems based on recent answers, but the suggestion mechanism is a difficulty-band rule, not a real adaptive engine. The state model is roughly "your current ability estimate on this topic," which is a single scalar updated through maximum-likelihood. That's less sophisticated than what ASSISTments was doing in 2010 with Bayesian Knowledge Tracing.
The reason this matters is not that Pearson is uniquely behind the curve. The reason it matters is that Pearson has publicly committed to an adaptive-learning strategy and has had three years to deliver on it. The fact that what's shipped is essentially a difficulty-band recommender layered on top of a digital textbook is evidence that Pearson's engineering organization isn't structured to build the thing they're promising. They're a publisher with a software team, not a software company with publishing assets.
The strategic vulnerability is acute. Pearson's primary value proposition to institutions is content — the textbooks, the assessment items, the curriculum-aligned material. AI synthesis has collapsed the cost of content by two and a half orders of magnitude, which means Pearson's core asset is in the process of being commoditized. Their pivot to "adaptive learning" is the right strategic instinct — moving up the value chain into the engine layer — but they don't have the engineering organization to build it. The result is going to be that Pearson keeps shipping Pearson+ as an evolutionary improvement on their textbook stack while AI-native competitors build the actual adaptive layer on top of commoditized content. Pearson is structurally going to lose this race unless they acquire the engine layer, and the longer they wait the more expensive the acquisition becomes.
Coursera
Coursera is the largest online course marketplace, with around 142 million registered learners and a revenue run rate around $750M. They've also been talking about adaptive learning for years, mostly through their "Coursera Coach" AI tutor product launched in 2023.
What Coursera ships is video courses with quizzes attached and a recommendation engine that suggests "what to take next" based on broad cohort patterns ("learners like you took these courses"). The Coursera Coach AI tutor is an LLM wrapper that answers questions about the course material. It's a useful feature — having an AI tutor that can explain a confusing point is genuinely valuable — but it's not adaptive learning in the technical sense. There's no per-learner state model, no real-time updates to a proficiency vector, no decision impact on what the system shows next based on what the learner just demonstrated. The "adaptation" is at the recommendation-engine layer, not the content-delivery layer.
The structural problem at Coursera is different from Pearson's. Coursera's business model is the marketplace — connecting instructors to learners. The instructors own the course content; Coursera owns the distribution. Adaptive learning, done correctly, requires deep integration with the content: the system needs to know what each question is actually testing, what the prerequisite relationships are, and how to map a learner's failure on question A to a knowledge gap that affects questions B, C, and D. None of that is possible at the level of integration Coursera has with its instructors' content. The instructors ship MP4s and PDFs and quiz JSON; Coursera serves them. There's no semantic understanding of what's in the content, and there's no architectural place to put one.
Coursera's strategic vulnerability is that the marketplace model assumes content is the scarce resource. When content is commoditized, marketplaces lose value relative to platforms that own both the content and the engine. Coursera is trying to bridge this by adding the AI tutor as a thin "platform" layer over the marketplace, but the bridge is structurally too narrow. The AI tutor can explain the content; it can't adapt the experience to the learner. That's a different architectural commitment than Coursera has made.
Khan Academy
Khan Academy is the most interesting of the three, because they've made the most credible engineering effort at adaptive learning of any non-profit edtech organization. The Khan Academy "mastery learning" framework actually attempts per-skill mastery tracking, prerequisite mapping, and adaptive practice recommendation. It's the closest any of the household-name edtech brands gets to meeting the technical definition.
But it's still not there. The Khan Academy mastery model uses a discrete state machine: each skill has a mastery state (familiar, proficient, mastered, level-up) that updates in response to answers, with simple transition rules. The system is meaningfully more adaptive than Pearson+ or Coursera, but it's a 2010-era architecture. The state is coarse, the updates are batch-y, and the cross-skill transfer logic is hand-written rather than learned. Khan's Khanmigo AI tutor (launched 2023) is an LLM wrapper similar in spirit to Coursera Coach — useful for explaining concepts, not part of the adaptive engine.
The structural problem at Khan Academy is that they're a non-profit with a relatively small engineering team and a mandate that includes K-12 education access at global scale. They have neither the engineering bandwidth nor the financial structure to rebuild the mastery engine on a modern adaptive substrate. What they have is the best educational data set in the consumer edtech world (billions of learner interactions across millions of students) and a brand that is universally trusted. Those are real moats. But they don't translate into adaptive-engine leadership without an engineering investment they're unlikely to make.
Khan Academy's most likely future is to either partner with or be absorbed into a larger adaptive-learning effort — possibly inside a Big Tech company (Microsoft, Google, OpenAI) that has both the engineering capacity and the patience for a long-term non-profit-aligned product play. The pure-play independent path is hard for them.
The Gap
The three brands together — Pearson, Coursera, Khan Academy — represent maybe 200 million registered learners and several billion dollars of revenue across consumer, institutional, and corporate channels. None of them ships an adaptive engine that meets the technical bar. The reason isn't that they don't see the opportunity. The reason is that each one has a structural blocker:
- Pearson: a publishing organization without the engineering DNA to build a modern adaptive substrate.
- Coursera: a marketplace whose business model precludes the deep content integration adaptive learning requires.
- Khan Academy: a non-profit with the most credible engineering attempt of the three, but constrained by mission scope and engineering bandwidth.
This is the gap that the next generation of edtech is being built into. The combination of (a) AI synthesis collapsing the cost of content, (b) real-time inference enabling true per-learner state updates, and (c) modern ML systems engineering being broadly available means that the technical bar for building a credible adaptive engine has fallen dramatically. The vendors who build now, on the right substrate, with the right team — primarily engineering-led, content-cheap, latency-first — get to define the next decade.
Who Wins
I'd bet on three classes of organization to define adaptive learning over the next five to ten years:
- AI-native edtech startups that built the adaptive engine from day one as a real-time ML system, with content as a generated artifact rather than a hand-authored input. This is the structural disposition that the legacy vendors can't easily retrofit.
- Big Tech with education adjacencies: Microsoft (LinkedIn Learning, GitHub Copilot's educational angle), Google (Workspace for Education, Gemini's tutoring features), OpenAI (ChatGPT for Education, which is structurally underbuilt today but could be rebuilt as a real adaptive engine). These companies have the engineering capacity to build correctly, but they have to choose to do it — and so far they've mostly been doing thin LLM-wrapper plays rather than full adaptive engines.
- Specific verticals where the existing incumbents are particularly weak: Wolters Kluwer's CE verticals are wide open; KnowBe4 is uniquely vulnerable in security awareness; A Cloud Guru / Pluralsight is mid-pivot in cert prep; the test-prep market (Kaplan, Princeton Review) is structurally exposed. A vertical-specific adaptive-learning vendor can dominate any one of these with a focused execution.
The names above — Pearson, Coursera, Khan Academy — are not on this list. That's not because they're bad companies. It's because the architectural choices that got each of them to their current scale make the pivot to true adaptive learning extremely expensive, and none of them has the disposition to make it. The category is going to consolidate around new entrants, with the legacy brands either acquired into the new platforms or relegated to distribution-layer roles.
The Honest Version
I'm a competitor to all three of these companies, so reasonable people might discount this analysis as motivated. The simple test is: try them yourself. Pick an exam you don't know well. Take the same diagnostic test on Pearson+, Coursera, and Khan Academy. Get a question wrong on each one. Then check what each system shows you next.
The Pearson and Coursera answers will be functionally indistinguishable from a static-content platform: a slightly different difficulty band, an unrelated next topic, or just "try again." Khan Academy will do slightly better, especially on math, because their mastery engine is more sophisticated than the other two — but the next question still won't reflect why you got the previous one wrong. It will reflect that you got it wrong, at a coarse level.
A real adaptive platform — one that meets the four technical criteria I described in the earlier post — should show you a different next question depending on which wrong answer you chose, how confident you were, and what prerequisite the failure implies. That's the gap. The next generation of platforms is going to close it. The brands above are not, structurally, going to be among them.
Part of an ongoing series on the engineering and economics of adaptive learning systems. Renkara Media Group's adaptive engine is protected by US patent applications — see the patent portfolio overview.