Google on the Offensive: Is This the End of AI EdTech?

Panic mode in the startup world as trillion-dollar companies come for their sandwich.

Lately, I had been feeling that those standard AI presentation disclaimers—”what we’re discussing might be obsolete by the time we finish”—were becoming a bit kitsch.

I was wrong.

This is the third or fourth rewrite of an article I started in September. After largely unplugging for the summer, and battling a prolonged fit of procrastinitis in the early fall, I finally sat down to write. But every time I did, new developments emerged that demanded I add on to what I was saying.

The picture that has finally come into focus is this: After a very awkward start in 2023, Google is aggressively taking the lead. They are giving the competition headaches, but more importantly for us, they are actively cannibalizing the AI EdTech business model. Will the Kahnmigos, MagicSchools, or SchoolAIs of the world survive the latest developments? That is remarkably unclear to me.

I won’t go too deep into the Gemini vs. ChatGPT vs. The World debate. Instead, I’ll focus on what these moves mean for the EdTech industry. Suffice it to say that 48 hours ago, Google released Gemini 3—available even in the free tier. After the somewhat tepid release of GPT-5.1 last week, this one is making a big splash, leapfrogging all other models substantially on the benchmarks.

The Giants Enter the Arena

Right around the end of the last school year, on June 30, Google announced that Gemini was not only accessible from Google Workspace Education accounts but, crucially, would be integrated directly into Google Classroom.

A month later, OpenAI released ChatGPT “Study Mode” to all subscribers, including free accounts. Google followed suit a week later with its own “Guided Learning” mode.

Why is this interesting?

Up until now, the big AI labs had not made significant moves to penetrate the EdTech market directly. Sure, OpenAI released a hybrid version called ChatGPT Edu last year, but that was largely a market-share play for Higher Ed—a way to get college students hooked on their ecosystem rather than the competition’s. It didn’t bring any radically new functionalities to the table.

This left the field wide open for a million and a half startups. These companies built “wrapper apps” leveraging chatbots via API, competing to streamline teacher workflows while gesturing vehemently toward students, arguing ad nauseam that they had the best pedagogical tool to “adapt to learners,” “personalize instruction,” and “empower kids” (yada, yada, yada).

ChatGPT Study Mode and Google’s Guided Learning represent a shift. The giants are now entering the pedagogical space with tools purporting to help users learn, supposedly like a teacher would. While we’re definitely not at a teacher-replacement point, they are stepping squarely onto AI EdTech turf. The custom AI-facing bots now have competition from companies with infinitely deeper pockets.

(I should note that Anthropic was actually the first AI lab to provide an education-specific tool with Claude for Education in April 2025, though not for free. It makes OpenAI’s release of ChatGPT Edu a month later look much more reactive in retrospect.)

But let’s talk about the most hostile turf takeover we’ve seen yet: the integration of Gemini within Google Classroom.

Gemini in the Classroom

https://classroom.google.com/ai

If you look at the Gemini integration, you’ll see a choice board of tools that looks suspiciously like the dashboards of SchoolAI or MagicSchool. However, there is a notable difference. Look at the icons on the top grid: some have a pictogram, others just have the Gemini blue star. The tools with the pictogram are essentially exactly like the MagicSchool tools: you’re brought to a page with textboxes and dropdowns where you prompt for grade level, subject, and standards. It is prompt engineering via form-filling.

On the other hand, if you try one of the tools with the blue star, the pre-programmed prompting isn’t even hidden anymore. A new tab opens with Gemini, and the pre-prompt is right there for you to see. If you’ve read my previous articles explaining how AI EdTech tools work and still had doubts, try these. It makes the “wrapper” process completely transparent, showing you how the sausage is made.

But where Google blows the competition out of the water is happening right below that grid, in the “Explore more Google AI Tools” section.

Brainstorm and Create with the Gemini App

Two days ago, I was writing that Gemini 2.5 Pro was my favorite model—so much so that last month, I broke up my almost 3-year relationship with ChatGPT Plus to give my hard-earned money to Google instead.

Well, as of November 18, 2025, all teachers in Google districts have free access to Gemini 3 Flash and Pro. That is, the state-of-the-art model, period.

If you’re thinking “big deal” because you’ve just been using chatbots to generate emails or rubrics, I understand the sentiment. But that’s because you’re stuck in 2023. You don’t know the Canvas mode of Gemini - for instance.

• Click HERE to use a math simulation on the Pythagorean theorem that I made in under 5 minutes.

• Click HERE to use a simulator I made even quicker to study the influence of mass and volume on buoyancy.

And finally, click THIS LINK (left/right arrows to move, space to jump) to play my Mario-style 7th-grade math game with a word problem final boss.

“Make a Nintendo-like game with mystery question boxes that the character can bump its head into like Mario... when you bump into a mystery box a question appears with a sound effect, and you need to answer some math questions 7th grade level to go on... after going on and answering about 5 questions... you get to the final boss which is a word problem on the same unit. If you answer one time wrong, you get a clue, if you miss again, game over. Before making the game why don’t you suggest improvements to my prompt to make it engaging to 7th grade students, but above all, to impress all my colleagues ;-)”

That one took me a bit longer: 37 minutes total. I persevered through 12 dysfunctional versions (about one generated every 3 minutes) to get this final product. I just complained and complained while Gemini patiently wrote 1,000 lines of code again and again.

If you can’t play it, don’t worry, I made a gif so you can see what it looks like.

NB: I failed on purpose at the last question, I swear!

No EdTech product on the market comes even close to this kind of capability, for any amount of money. And if you keep reading, you might surmise that it may never be made available to them.

The Diamond on the Cherry on the Icing

If that wasn’t enough, you get to use Nano Banana for image creation. (If you don’t know what that is, urgently read my article on the topic).

Look, I made myself look young again:

And just to show off, I asked Nano Banana PRO (released today) to take this picture and make a comic strip of me and French clichés in the style of Uderzo.

Personalize Gemini: Create a Gem

You also get to make Gems, which are the Gemini version of ChatGPT’s custom GPTs. If you haven’t been using Custom GPTs, Gems, or Claude Projects, this is the next step in leveraging AI for productivity.

As I explained in my long-term prediction article: “Your custom GPTs are like a private army of employees... A ‘what are we eating for dinner’ bot... A ‘report to my boss’ bot... they all have specific personalities, knowledge bases, and goals.”

Every teacher person should have a Curriculum Vitae Gem with all your best resumes and cover letters.

For classroom teachers: while I exhort you not to share PII with any AI that isn’t sandboxed and approved, nothing prevents you from communicating an accurate picture of your class using a code system (Student 7A, Student 7B, etc.). If your district provides Gemini access, and the code key isn’t shared, you can create a Gem specific to your roster or unit with supporting documents. It is incredibly helpful. Ask your Data Privacy Officer.

Grow Understanding: Create a Notebook

That’s right, you now have access to NotebookLM.

You’ve likely heard about the podcast maker. While the customization isn’t quite where I’d like it to be for instructional materials, it is still extremely useful. You can now create mp4 slideshows, three formats of podcasts, quizzes, flashcards, and mind maps. (See a demo here).

I use it to “read” science—transforming papers into mp3s I listen to while driving. It’s not text-to-speech; it’s a debate between two AI characters who sound like they work at NPR. For UDL implementation, providing content in different formats with a few clicks is a game changer. No EdTech tool has anything like this.

Market Penetration: The Numbers Game

So, Gemini is free for teachers and includes tools that paid EdTech tiers don’t have. But the real killer is the distribution.

Source

I’m not going to go full rabbit hole here, but market shares are hard to compare. MagicSchool claims partnerships with 10,000 schools; SchoolAI works with “hundreds of districts.” That’s great for them.

But Google Classroom is used in 60,000 schools in the US alone and is present in 230 countries¹. Depending on the source, they hold 40 to 50% market share in K12.

You see where I’m going with this?

Google entering the space cuts out the middleman. With investment capabilities dwarfing even the most enthusiastic VCs, they are pulverizing the competition’s market penetration. The value add of EdTech wrapper apps rests on educators designing tools to simplify workflows. Google can simply poach those educators.

I suspect some EdTech founders and CEOs are currently hoping to get bought up, rather than just pushed out.

(Side note: OpenAI signed a partnership with Canvas in July. It seems targeted at Higher Ed, but you can watch the demo video here if you have 20 minutes, it left me rather nonplussed).

Built-in Pedagogy

Let’s abandon the teacher-facing side for a moment. As of November 2025, Gemini ranks numero uno in what I believe to be the best benchmark.

But is it a good substitute for a teacher? Can the Guided Learning mode help you achieve true understanding of something you don’t know?

I tested it. Instead of impersonating a student, I dug into my past to find a concept I genuinely gave up on: matrix diagonalization. I used Gemini Guided Learning to try to learn it.

And... I succeeded. Don’t quiz me, but I remember the eigenvalues, eigenvectors, the determinant, and more importantly: why anyone would want to do that in the first place.

So, it works?

Not so fast.

We know learning gains can occur with AI if motivation is present. The Harvard study showed physics undergrads outperforming peers using a custom bot. But Harvard undergrads are not representative of the general K12 population.

Just like with Intelligent Tutoring Systems (the “5% problem”), the tech works for students who use it “as intended”—those who are motivated and have good metacognition. The interaction is still almost entirely text-based. While that worked for me, it’s not going to engage little Timmy who has given up on Algebra. Visuals are arriving fast though, this paragraph will age fast.

However, if Google is filling the AI tutor space, this is where we will see the next significant advances. And I have news on that front.

The Game Changer: Nested Learning

In my previous articles, I identified two major weaknesses of current AI tutors that effectively prevent them from replacing human teachers. Both come down to memory.

1. Anterograde Amnesia: This isn’t my metaphor; it’s the literal diagnosis given by Google researchers in a fresh new paper. Current AI models reboot every time a chat opens. They cannot remember long-term context or shared history. Teaching is fundamentally about building relationships, and you cannot build a relationship with a colleague who introduces themselves to you every single morning and asks what the job is all about.

2. Static Knowledge (The Parrot Problem): You can feed a bot every pedagogy book in the world via RAG, and it will become a savant parrot—retrieving facts, but not learning from them. It doesn’t get better at teaching the more it teaches; it just resets.

Earlier this week, I learned about a new paper by Google engineers that validates this analysis—and proposes a fix.

They call this new learning paradigm Nested Learning (NL), and they have built a specific self-modifying architecture called HOPE. Explicitly inspired by neuroplasticity, it is designed to cure that “anterograde amnesia.”

Here is the gist of what this new architecture achieves:

“The HOPE architecture represents a functional upgrade over the current generation of AI... First, it effectively cures ‘anterograde amnesia’ through a sophisticated Continuum Memory System (CMS). This gives the AI longitudinal memory, allowing it to find patterns buried deep in a student’s history. Second, and more revolutionary, the self-modifying nature of HOPE allows the AI to internalize and improve its skills over time. It can actually acquire new pedagogical approaches without overwriting its fundamental knowledge base.”

In plain English? Google is creating a system designed to be the opposite of your amnesic colleague: a coherent, adaptive learning module that improves how it teaches based on who it’s teaching.

If Google integrates this into their free, ubiquitous ecosystem, the “wrapper app” era won’t just be disrupted—it will be obsolete.

Conclusion: The Squeeze

Pheww... that was a lot of new news. Thanks for making it this far.

Before we go, shall we have a little recap? If you haven’t been glued to the AI news feeds this past few months, here is the “Too Long; Didn’t Read” of the current landscape:

• The “Plateau” Myth: For all the talk in late 2024 about LLMs hitting a wall, the recent releases of Gemini 3 and GPT-5.1 proved that the ceiling is still rising—it’s just rising inside the fortresses of Trillion-dollar companies.

• Google is Everywhere: With Gemini integrated directly into Classroom, Workspace, and now offering “Guided Learning,” they have effectively commoditized the features that most EdTech startups charge for.

• The “Reverse Acquihire”: This is the Silicon Valley trend of 2025. Big Tech isn’t buying startups; they are hiring the founders, licensing the tech, and leaving the company as a shell to avoid antitrust scrutiny.

So, where does this leave the industry?

In my previous prediction article, I suggested that the smartest play for survival would be a merger between established Intelligent Tutoring Systems (ITS) and GenAI startups. The logic is simple: ITS platforms like IXL have the validity and the data, but they have the “5% problem”—kids don’t want to use them. GenAI tools are promising for engagement, but they lack validity. Smash them together, and you get a conversational tool that is both engaging and accurate.

We are already seeing the first clumsy steps in this direction. IXL released Spark Studio, their own GenAI platform. Crucially, however, it is currently separate from their core ITS product and does not interact with student data. It’s an island of GenAI floating next to the continent of traditional practice. The winner will be the one who builds the bridge.

If they don’t build that bridge, I see only two other paths for these founders, neither of which is particularly “fun.” They might aim for a “Reverse Acquihire,” hoping to get poached by Microsoft or Amazon while their product is left to drift into obsolescence—just like we saw with Inflection, Adept, and Character.ai. Or, they might retreat to “The Hiding Spot”—finding a niche so boring, bureaucratic, or high-liability that Google simply can’t be bothered to automate it. If you can build a moat around a workflow that is too annoying for Sam Altman to care about, you might just survive.

For everyone else? The giants are here. And they aren’t just eating your lunch; they’re the ones who packed your lunchbox in the first place.

1. That tripped me too, aren’t there just 195 countries? Look at the source, I think it says “countries and territories” https://electroiq.com/stats/google-classroom-statistics/