What are the latest news in the AI in education market?

What's actually new in AI education right now?

AI education is moving into a more serious phase, with new classroom products arriving almost every month and fresh research raising uncomfortable questions about whether students are learning more.

Some of the most interesting developments have happened lately.

On October 14, Gallup published a survey showing that 59% of American K-12 educators used AI professionally during the previous school year. A week earlier, OpenAI reported that nearly 1.2 million teenagers had used ChatGPT's Learning Visualizations in a single week.

Meanwhile, Google expanded its partnership with Khan Academy in August, Anthropic launched Claude for Teachers in July, and OpenAI introduced education-specific plugins for students and educators in August.

But the most surprising news came from university research. A newly published randomized experiment found that giving students access to an AI tutor actually lowered their grades and reduced their participation in existing learning activities.

These developments point in two directions. AI is becoming a normal part of education, and the companies building it are making increasingly sophisticated products. Yet we still have serious questions about what happens to students when AI takes over too much of their work.

That gap between widespread adoption and actual learning outcomes is becoming the most important issue in AI education today.

Has AI use become almost universal at universities?

AI use is now nearly universal among British university students, and the increase over the past two years has been remarkable.

The Higher Education Policy Institute's March 2026 survey found that 95% of UK undergraduates had used AI tools, compared with 92% in 2025 and 66% in 2024.

The change becomes even more striking when we look specifically at assessed work.

In 2024, 53% of students said they had used generative AI to help prepare assignments. That increased to 88% in 2025 and 94% in 2026.

Within two years, AI assistance moved from something roughly half of students had tried to something almost everyone had used.

Most of that activity involves explanations, summaries and research assistance. Some 61% of respondents had used AI to understand concepts, 49% to summarize articles and 40% to develop research ideas.

The share directly incorporating unedited AI-generated text into assessed work was much smaller, at 12%.

There is also an interesting development beneath the overall adoption figures. Although more students are using AI, the proportion using it specifically to generate text fell from 64% in 2025 to 56% in 2026, according to the report.

Students appear to be spreading their activity across more specialized tools, including summarization, note-taking, quizzes and writing assistance. The survey does not establish exactly why text generation declined, but overall AI adoption clearly hasn't slowed.

The latest study covered 1,054 UK undergraduates, so we shouldn't apply these percentages to every country. Still, the three consecutive surveys give us a clear picture of how quickly university behavior has changed.

AI use among UK undergraduates	2024	2025	2026
Used AI in any way	66%	92%	95%
Used AI for assessed work	53%	88%	94%
Used AI to explain concepts	36%	58%	61%
Used AI to summarize articles	24%	47%	49%
Used AI for research ideas	25%	40%	40%
Generated text and edited it	13%	25%	25%
Included AI text without editing	3%	8%	12%

How many teenagers already rely on AI for homework?

More than half of American teenagers now use AI chatbots for schoolwork, although only a minority depend on them for most assignments.

Pew Research Center's February 2026 report found that 54% of teenagers aged 13 to 17 had used chatbots for schoolwork, while 57% had used them to search for information.

Around three in ten teenagers reported using AI chatbots daily, across all activities.

Pew also asked how much schoolwork students completed with AI assistance. Some 10% said chatbots helped with all or most of their schoolwork, 21% said some, and another 23% said a little.

So we have a growing group that relies heavily on AI, alongside a much larger group that uses it occasionally.

OpenAI's own usage data adds another perspective.

In August 2026, the company reported that American schoolwork-related prompts to ChatGPT exceeded 460 million messages per week during the academic year, compared with more than 180 million weekly during the summer.

Those are message counts rather than unique users. Nevertheless, the strong seasonal pattern is revealing. Students turn to AI much more frequently when school is in session.

The company's newer ChatGPT for Teens experience reflects this demand, offering guided explanations, quizzes, study planning and homework reminders.

OpenAI also reported that more than 180,000 teenagers used Study Mode in one week. These figures establish meaningful usage, although they tell us little about whether the students subsequently performed better in school.

The strongest conclusion from the latest numbers is that AI-assisted homework has become ordinary behavior for American teenagers, with a smaller but significant group using it for most of their academic work.

Are teachers actually saving time with AI?

Teachers who use AI are generally reporting real time savings, and the benefits are much clearer for lesson preparation than for replacing classroom instruction.

Gallup's October 2026 research surveyed 1,952 American K-12 educators, including teachers, paraprofessionals and other classroom staff.

Among educators who had used AI during the previous school year, 64% said it freed up more time for the activities they valued most. Only 7% said it cost them time.

An earlier Gallup-Walton Family Foundation study offers a useful estimate of the practical benefit. Teachers who used AI at least weekly reported saving an average of 5.9 hours per week.

Across a 36-week school year, that would represent approximately 212 hours per teacher if the same weekly savings continued. It's an illustrative calculation from self-reported time savings, rather than a measured annual total.

The work being automated is also quite specific.

Teachers use AI to create lesson plans, adapt reading materials, generate worksheets, prepare assessments and handle administrative tasks. These are activities that previously required substantial preparation outside classroom hours.

Gallup's latest survey found that 85% of educators felt they mattered a great deal when helping a student through something difficult, compared with much weaker feelings about administrative work.

This helps explain why teacher-focused AI products have attracted so much attention.

MagicSchool specializes in lesson preparation and instructional resources. Microsoft's Teach generates lesson plans, quizzes and grading rubrics. Claude for Teachers provides access to teaching tools connected to educational standards.

All three concentrate on helping teachers do their existing jobs more efficiently.

For now, we found much stronger evidence that AI is reducing teachers' preparation workload than that schools are replacing teachers with AI.

Does AI actually help students learn more?

AI tutoring can produce impressive learning gains, but those results depend heavily on how the lessons are designed.

One of the strongest positive findings comes from Harvard University.

In a randomized study published in Scientific Reports in June 2025, researchers compared a carefully designed AI tutor with conventional active-learning instruction among 194 undergraduate physics students.

Students using the AI tutor achieved more than twice the median learning gain of students attending the conventional lessons. They also completed the learning activities in less time and reported higher engagement.

The result deserves attention because the researchers compared actual learning outcomes rather than simply asking students whether they liked the technology.

However, the tutor had been specifically designed for the course. Researchers used teaching expertise and educational psychology to structure explanations, questions and feedback.

The study also covered two physics lessons, which limits what we can conclude about semester-long academic performance.

Other controlled experiments have found meaningful gains in different settings, including secondary education. But unrestricted chatbot access has sometimes produced worse outcomes when students must work independently afterward.

The evidence is therefore quite specific. A well-designed AI tutor can outperform conventional instruction in certain tasks, while a general-purpose chatbot can make students look more competent during practice than they really are.

That's a major difference for education companies promising better academic results.

If you want more recent data on this point, please see our latest AI in education market report.

Can AI tutors actually make university grades worse?

Yes, and one of the newest large-scale experiments found that students offered an AI tutor earned lower grades than comparable students without access.

Researchers at the University of Maryland published the findings in October 2026.

Their randomized trial involved 2,379 undergraduates and 30 instructors across several subjects during the fall 2025 semester.

In comparisons between sections of the same course, students offered the tutor finished with grades approximately four percentage points lower than their peers.

The estimated academic effect was a decline of 0.37 standard deviations.

Even more striking, participation in the university's learning management system dropped by 0.90 standard deviations among students given access to the tutor.

Researchers observed fewer interactions with existing learning materials, including reductions in page views and active days.

One possible explanation is that some students relied on the AI assistant instead of engaging with coursework that would have helped them learn.

The researchers also found larger estimated academic losses among first-generation college students.

We should distinguish the strength of the finding from how broadly it applies. The study is a working paper based on one public university and one implementation. It has not established that every AI tutor will produce negative results.

Still, the experiment is unusually important because it examines thousands of students, multiple subjects and final course grades.

Schools evaluating AI tutors now have a concrete reason to demand independent evidence before assuming that adding an assistant will improve academic performance.

Why do some AI tutors make students learn less?

AI tutors can weaken learning when they make it too easy to finish an assignment without understanding how to solve the problem independently.

A University of Pennsylvania-led study gives us particularly useful evidence because researchers compared two versions of GPT-4-based tutoring.

The experiment involved nearly 1,000 high school mathematics students in Turkey during the 2023–2024 academic year, with findings published in Proceedings of the National Academy of Sciences in 2025.

One version behaved more like an unrestricted chatbot. The other was designed to provide hints, use teacher-approved materials and encourage students to work through problems.

Both versions helped students complete practice exercises.

During AI-assisted sessions, performance increased by 48% with the unrestricted version and 127% with the structured tutor, compared with students without AI assistance.

But when researchers removed the tools and tested students independently, the unrestricted group performed approximately 17% worse than the control group.

The structured tutor group performed about as well as students who hadn't used AI.

We found this comparison especially convincing because both systems used the same underlying generation of AI technology.

What changed was the tutoring approach.

Students using unrestricted assistance could obtain answers quickly. The guided tutor encouraged them to attempt the mathematics and understand intermediate steps.

The difference also helps explain why newer products increasingly include guided questions, staged explanations and checks for understanding.

AI companies can make their products more useful educationally by changing how students interact with them, even without changing the underlying model.

If you want more recent data on this point, please see our latest AI in education market report.

Can AI really improve education in poorer countries?

AI tutoring has already delivered meaningful results in lower-resource schools, particularly when students receive structured guidance from teachers.

A World Bank-supported experiment in Edo State, Nigeria, provides one of the clearest examples.

Secondary school students joined a six-week after-school program combining Microsoft Copilot, powered by GPT-4, with teacher support.

Researchers conducted a randomized controlled trial and measured improvements in English, AI knowledge and digital skills.

Students participating in the program achieved an overall improvement of 0.31 standard deviations. The effect specifically on English was 0.23 standard deviations.

Those are substantial results for a short intervention.

Using a benchmark based on typical learning progress, the World Bank estimated that the improvement was comparable to approximately 1.5 to 2 years of conventional schooling.

That comparison requires care. The researchers measured gains on particular assessments, rather than demonstrating that six weeks of tutoring could replace two entire school years.

The teaching model is equally important.

Students attended organized sessions with human supervision and a defined curriculum. The program was much closer to structured tutoring than independent chatbot use.

The World Bank has also been exploring teacher-centered AI programs in countries including Peru and Brazil, suggesting that the Nigerian experiment is part of a broader effort to test practical educational applications.

At the national level, Estonia's AI Leap initiative provides another example of serious investment. Its original plan covered 20,000 secondary students and 3,000 teachers, with a subsequent phase targeting another 38,000 students and 2,000 teachers.

These are rollout targets rather than independently measured educational gains.

The Nigerian experiment is stronger evidence that AI-assisted teaching can help students learn. Estonia shows how governments are trying to make access and teacher training available on a much larger scale.

Are Google, OpenAI and Anthropic taking over AI education?

Google, OpenAI, Anthropic and Microsoft are becoming major providers of educational AI, and their products are increasingly built around how schools already operate.

Google currently has one of the clearest distribution advantages.

In September 2025, it reported that Gemini for Education had been integrated into more than 1,000 American higher education institutions, giving over 10 million students access.

Those figures measure institutional reach rather than active usage by every student. Still, they show how quickly Google can distribute AI through existing educational relationships.

Google also made more than 30 Gemini-powered Classroom tools available at no additional cost to eligible educators.

Microsoft has taken a similar approach by incorporating AI lesson planning, quizzes and assessments into Microsoft 365 Education.

Anthropic entered the teacher market more directly in July 2026 with free premium Claude access for verified American K-12 educators, followed by a school and district offering in August.

OpenAI, meanwhile, launched three education-specific plugins in August covering K-12 teachers, university educators and college students.

The more interesting recent development concerns personalization.

Google's work with Khan Academy now includes interactive visual diagrams that students can manipulate during mathematical explanations. OpenAI offers guided learning, quizzes, flashcards and visualizations. Microsoft is connecting study assistance with course materials.

We reviewed the major launches from these providers and found a consistent move toward tools that understand assignments, learning materials and classroom workflows.

That puts additional pressure on smaller companies selling relatively simple AI homework assistance.

Company	Education product or expansion	Announcement	Main feature
Google	Gemini in Classroom	June 2025	30+ AI teaching tools at no additional cost
Google	Gemini for Education	September 2025 milestone	Access across 1,000+ US colleges
Microsoft	Teach in Microsoft 365	January 2026 expansion	Lesson plans, assessments and rubrics
Google + Khan Academy	Gemini tutoring partnership	January 2026	Guided reading and writing support
Anthropic	Claude for Teachers	July 2026	Free premium AI for verified US teachers
OpenAI	Education plugins	August 2026	Dedicated workflows for students and educators
OpenAI	ChatGPT for Teens	August 2026	Guided homework, quizzes and study controls
Google + Khan Academy	New Khanmigo capabilities	August 2026	Interactive diagrams and teacher-controlled practice
Anthropic	School and district offering	August 2026	Institutional administration and data controls

Can smaller AI education startups survive against ChatGPT?

Specialized AI education companies can still build large businesses, but their products need to solve problems that a general chatbot handles poorly.

MagicSchool offers one of the best examples.

The company, founded in 2023, focuses on practical teaching tasks such as curriculum planning, differentiated instruction, feedback and classroom documentation.

An Anthropic case study reported that MagicSchool supported 3 million educators, more than 100 million AI-powered interactions and over 5,500 school and district partnerships.

The same case study reported that approximately 1 million students had been reached through teacher-guided AI interactions.

These company-reported numbers don't establish how much revenue MagicSchool makes, but the distribution is significant for a relatively young company.

Khan Academy represents a different strategy.

Its AI tutoring product has existed since the original Khanmigo pilot in 2023, initially supported by OpenAI technology. Khan Academy has since expanded its relationship with Google and introduced new Gemini-powered learning experiences.

The company already has established educational content, teachers who know its products and relationships with schools.

Those advantages remain valuable even as the underlying AI provider changes.

We can see why this matters commercially by comparing the companies.

MagicSchool concentrates on teacher workflows and school adoption. Khan Academy brings its educational materials and established learning environment. Both can incorporate increasingly capable AI models without needing to build those models themselves.

A startup offering little more than a chatbot with homework-themed branding faces a much harder market.

Its features can be copied by general-purpose AI products with much larger distribution.

The strongest independent providers are likely to be those that understand curriculum requirements, school purchasing, classroom controls and the everyday problems educators actually face.

If you want more recent data on this point, please see our latest AI in education market report.

Are universities falling behind students on AI?

Universities are still struggling to keep up with how widely students use AI, leaving many learners to choose their own tools without much institutional support.

The latest Higher Education Policy Institute survey makes the gap unusually clear.

Some 95% of British undergraduates had used AI, yet only 36% felt encouraged by their university to use it. Just 38% said their institution provided AI tools.

That leaves a 57-percentage-point gap between overall student usage and institution-provided access.

The difference doesn't necessarily mean universities are deliberately resisting AI.

Students can access free versions of ChatGPT, Gemini and other services independently, often before their university has decided which systems to support.

But independent adoption creates problems for universities.

Some students use institution-approved products with clearer privacy protections, while others rely on personal accounts. Teachers may struggle to establish consistent rules, and universities have limited visibility into the tools their students depend on.

Institutional encouragement has improved, rising from 29% in 2025 to 36% in 2026.

However, student usage has grown much faster over the same period.

This situation creates a commercial opportunity for companies offering managed educational AI.

Universities have reasons to pay for administration, privacy controls, course integration and consistent access even when students can already use similar AI capabilities for free.

The question for vendors is whether those additional services are valuable enough to justify institutional budgets.

Is AI making cheating much more common in schools?

AI has made academic misconduct easier, and recent surveys show that direct use of generated text is increasing, although the actual cheating rate remains difficult to measure.

Among British undergraduates, the proportion submitting AI-generated text without editing rose fourfold between 2024 and 2026, as shown earlier.

That is a meaningful behavioral change, but it doesn't tell us precisely how many students violated their university's rules.

Some courses explicitly allow generated text. Others prohibit assistance that might appear relatively harmless elsewhere.

Pew Research Center asked American teenagers a different question in its February 2026 survey.

Some 59% believed students at their school used AI to cheat at least somewhat often. Approximately one-third believed it happened very or extremely often.

Among teenagers who personally used chatbots for schoolwork, 76% believed cheating occurred at least sometimes.

Those figures measure perceptions rather than verified misconduct. Students who use AI themselves may also be more aware of how classmates use it.

However, schools are clearly responding.

In the British university survey, 65% of students said their assessments had changed significantly because of AI.

That is arguably the strongest evidence of how disruptive AI has become for academic evaluation.

Teachers can no longer assume that a polished essay or correctly solved take-home problem demonstrates independent student ability.

The practical response is already taking shape through more supervised assessments, oral explanations, staged submissions and assignments where students must demonstrate how they arrived at their answers.

These changes are likely to outlast the current generation of AI tools because the underlying problem becomes harder as models improve.

Can AI detectors really tell when students cheated?

AI detectors remain too unreliable to prove that a student cheated, and even major detection companies warn schools against treating their results as definitive.

Turnitin provides a particularly useful example.

Its official guidance explains that AI writing detection can incorrectly classify human-written, AI-generated and AI-paraphrased text.

The company explicitly advises educators against using its detection scores as the sole basis for adverse action.

Turnitin has also changed how its software displays low detection scores.

For results between 1% and 19%, the company no longer displays a precise percentage or highlighted passages, partly because false positives are more likely in that range.

That is a significant limitation for a product used in academic integrity investigations.

The broader difficulty is that students can use AI in many different ways.

A student might write an original essay and use AI to improve grammar. Another might generate an entire assignment and rewrite enough passages to avoid detection.

The resulting text can be difficult to classify reliably, particularly when rules vary between institutions.

California's newly published model school AI policy reflects these concerns. It includes provisions addressing the limitations of AI detection software and the need for human judgment.

Detection tools may help identify submissions worth reviewing, but schools still need independent evidence before deciding that misconduct occurred.

Trying to solve the entire cheating problem through software detection looks increasingly unrealistic.

If you want more recent data on this point, please see our latest AI in education market report.

Are governments finally putting rules around AI in schools?

Governments are starting to introduce clearer rules for educational AI, particularly around privacy, academic integrity and automated decisions affecting students.

California released a model policy for K-12 education on June 25, 2026, followed by an official announcement on July 6.

The policy gives school districts a framework covering responsible AI use, student data, disclosure, teacher judgment and academic integrity.

It also addresses AI literacy and parents' rights.

Importantly, the framework is voluntary guidance that local educational agencies can adapt. California has not imposed one identical AI policy on every school through this document.

Europe is moving further toward legally enforceable obligations.

Under the EU AI Act, certain educational applications are classified as high-risk when they influence decisions such as admissions, academic evaluation or assigning students to educational levels.

The current implementation timetable sets December 2, 2027, for applying the relevant high-risk requirements to Annex III systems.

The distinction is important for developers.

A chatbot helping students understand algebra generally faces different regulatory considerations from an automated system deciding whether a student qualifies for admission.

Privacy is another issue, especially for younger students.

In August 2026, Anthropic clarified that identifiable student information should only be processed through its education offerings with appropriate authorization from schools or districts.

These developments suggest that AI education providers will increasingly need to satisfy procurement, safety and data-handling requirements.

Companies already working closely with school administrators may find those requirements easier to meet than consumer chatbot providers entering the institutional market.

Is investor money still pouring into AI education?

Investors are still funding AI education companies, but overall EdTech investment is far below its earlier peak and has weakened again recently.

HolonIQ recorded approximately $1 billion in global EdTech venture funding during the first half of 2026.

That was 26% below the $1.35 billion raised during the first half of 2025.

Deal volumes remained relatively stable, according to HolonIQ, suggesting that investors were writing smaller checks rather than abandoning the sector altogether.

The longer-term comparison is even more revealing.

Global EdTech venture investment reached $20.8 billion in 2021, according to HolonIQ's historical data.

By 2025, annual funding had fallen to approximately $2.6 billion.

That means the industry was attracting only about one-eighth of the venture capital it received at its 2021 peak.

The latest figures therefore reflect a very different financing environment from the pandemic-era boom.

Yet some large deals continue.

Preply raised $150 million in January 2026 at a $1.2 billion valuation. Multiverse secured a $70 million round during the first half, while AI-enabled learning providers including Subject and Gizmo also attracted funding.

These companies operate in different parts of education, from language tutoring to workforce training and personalized K-12 learning.

We should also avoid confusing total EdTech funding with money invested exclusively in AI-native startups. HolonIQ's category includes a much broader range of educational businesses.

Still, the contrast is striking. AI usage among students and educators continues to increase while venture investment in the overall education technology industry remains subdued.

Investors are becoming more selective, with established traction and clear commercial demand carrying more weight than AI branding alone.

Which AI education companies are actually growing?

Duolingo and parts of Pearson are still growing, while Chegg is losing revenue at an extraordinary rate. The commercial results of education companies are becoming increasingly different.

Duolingo reported $298.5 million in second-quarter 2026 revenue, an 18% increase from the previous year.

Daily active users grew 23%, and the company reported a 72.6% gross margin.

Its expanding AI features, including Video Call, are part of a broader strategy to improve the learning experience.

However, we cannot attribute all of Duolingo's growth to AI. Its brand, distribution and existing subscription business remain important.

Pearson offers a second example.

The education company's first-half 2026 revenue reached £1.779 billion, up 4% on an underlying basis.

Its Virtual Learning division grew 19%, supported by higher enrollment, while Higher Education grew 2%.

Pearson is developing AI-powered learning products, although the financial results don't isolate how much revenue those features generated.

Chegg shows how dangerous the transition can be for older business models.

In August 2026, Chegg reported second-quarter revenue of $51.8 million, down 51% from $105.1 million a year earlier.

That represents more than $53 million in quarterly revenue lost in just twelve months.

Chegg has faced substantial pressure as students increasingly obtain homework explanations and academic assistance from general-purpose AI tools and AI-powered search.

The comparison with Duolingo is particularly telling.

During the same quarter, Duolingo generated approximately 5.8 times as much revenue as Chegg, while the two companies moved in completely opposite directions.

Preply adds another commercial model worth watching. The company connects more than 100,000 human tutors with learners across 180 countries and has incorporated AI into its teaching platform. TechCrunch reported that Preply had been EBITDA-profitable for twelve months when it announced its latest funding round.

We shouldn't treat these four businesses as direct competitors. They sell different educational services.

AI can strengthen companies with useful products and strong customer relationships, while putting enormous pressure on businesses whose main offering can increasingly be reproduced by a chatbot.

Will schools pay for AI when Google and Microsoft give it away?

Schools will still pay for specialized AI products, but basic content generation is becoming a very difficult feature to charge for.

Google already offers more than 30 AI tools in Classroom at no additional cost to eligible educators.

Microsoft provides Teach through Microsoft 365 Education without requiring a separate paid Copilot license.

Anthropic's individual teacher offering is also free for verified educators who meet its conditions.

These products cover much of what smaller educational AI startups initially sold: lesson plans, quizzes, worksheets, summaries and adapted reading materials.

This creates a real pricing problem.

If a teacher can generate a usable lesson plan through software the school already has, an independent provider needs to offer something meaningfully better.

We expect the stronger paid opportunities to involve deeper school-specific functionality.

That could mean curriculum alignment, assessment workflows, administrative controls, student progress tracking or tools that have demonstrated better learning outcomes in controlled studies.

There is already some evidence supporting this direction.

MagicSchool has built thousands of school and district relationships around specialized teaching workflows. Khan Academy is developing tutor experiences connected to its educational content and educator controls. Institution-managed offerings from the major AI providers include privacy and administration features absent from basic consumer accounts.

The commercial challenge is proving that those additional capabilities are worth paying for.

For now, AI education companies selling generic teaching assistance face growing pressure from free alternatives. Providers solving harder institutional problems have a more credible reason to charge.

If you want more recent data on this point, please see our latest AI in education market report.

So, what's the latest in AI education, and where is the market heading?

AI education is growing quickly, but the market is becoming much more demanding. Adoption is already widespread, while evidence about learning outcomes and commercial success remains surprisingly mixed.

Students are using AI for homework and university assignments at enormous scale. Teachers increasingly rely on it for lesson preparation, and some report saving several hours every week.

Major technology companies are also moving aggressively into education, bringing specialized tutoring features, classroom integrations and institution-managed tools to millions of potential users.

Yet the newest university experiments show that providing AI assistance can sometimes make students perform worse.

That finding is especially important because earlier controlled studies have demonstrated substantial benefits from carefully designed AI tutoring.

When we compare the research, the difference seems to come largely from how students use the technology and whether the system encourages them to work through problems themselves.

The commercial market is showing a similarly uneven picture. Some established education businesses continue growing, others are losing customers, and venture funding remains far below its former peak.

We therefore see a market that has largely passed the adoption test but is still working through the much harder question of educational value.

The next winners will probably be companies that can demonstrate better student outcomes, save educators meaningful time or solve specific problems for schools that general-purpose AI tools handle poorly.

For AI education, the current wave of product launches and user growth is impressive. But the companies that actually improve how people learn will have the strongest case for becoming lasting businesses.