How can a small edtech company start using AI without a big team?

Start with the one metric AI can move fastest, usually completion or retention. An adaptive path that adjusts difficulty per learner, or an AI tutor that answers questions inside the lesson, can be built on existing model APIs. You don't need to train your own model, you need clean content and one clear learning outcome to improve.

What are edtech companies actually using AI for in 2026?

Three things dominate: personalized and adaptive learning (Khan Academy's Khanmigo, Duolingo Max), AI content and assessment generation (turning source material into exercises and quizzes), and AI tutoring plus student support at scale. Khan Academy, Duolingo, and Coursera lead 5W's 2026 EdTech AI Visibility Index.

Should we build our own model or use an existing one?

Use an existing one. Almost every edtech AI feature that works in production runs on a general model API with your content and pedagogy on top. Building a custom model is rarely the bottleneck. Clean content, clear outcomes, and good prompts are.

How to Implement AI in an EdTech Business (3 Real Examples)

Q: Does an AI tutor actually improve learning outcomes?

It helps engagement, but the honest data is humbling. Khan Academy admitted only about 15 percent of students with access to Khanmigo actually use it, and that practice, not the AI itself, still drives the measured learning gains. Treat AI as a layer on top of a working product, not a replacement for it.

The AI that moves retention, not the AI that demos well

Most edtech teams already bolted a chat box onto their course and called it AI. It answers a few questions and it doesn’t move a single retention number.

The AI that changes an edtech business shows up in three places: how the content adapts to each learner, how fast you can produce and grade that content, and how well a student gets unstuck at 11pm without a human. The category leaders have proof for all three. A small edtech can copy the same playbook with a much smaller team.

Here are the three, with the companies running them and the honest limit most posts skip.

1. Adaptive learning: meet each student where they are

Khan Academy’s Khanmigo is the category-defining example. It grew from 68,000 pilot users in 2023-24 to over 1.4 million students by mid-2025, and it works by tutoring a student through a problem instead of handing over the answer. Duolingo pushed the same idea into language practice with Duolingo Max, using AI to explain mistakes and run roleplay conversations.

The mechanism that matters for an operator is personalization. A path that adjusts difficulty based on what a learner just got wrong keeps more people in the product than a fixed syllabus that bores half of them and loses the other half.

For a small edtech, this isn’t a research project. It’s a scoring layer plus a content map: tag your lessons by skill and difficulty, then let a model pick the next best exercise from what the learner just did. The inputs already sit in your course data.

2. Generate and grade content at scale

The slowest, most expensive part of running an edtech is producing good material. AI compresses it. Feed a model your source content and it drafts exercises, quizzes, and explanations, which a human then reviews. Auto-grading of open responses frees instructor time for the cases that actually need judgment.

Coursera built an AI Coach into its platform for exactly this kind of in-course support and feedback, and it sits third on 5W’s 2026 EdTech AI Visibility Index behind Khan Academy and Duolingo. The pattern scales down cleanly.

The guardrail here is review. A generated quiz question has to be checked for accuracy and for the sneaky failure where the model writes a question with no correct answer among the options. AI does the first draft and the volume, a human owns what ships. We wrote about where these agents break in production in what actually happens 30 days in.

3. Tutoring and support that answers instantly

Students churn when they get stuck and no one answers. An AI tutor embedded in the lesson closes that gap, walking a learner through the concept in the moment instead of pointing them to a forum. For the business side, the same instant-response logic qualifies and answers enrollment questions before a prospect cools off.

This is usually the fastest win because the value is immediate and measurable: track how many support questions resolve without a human, and how many trial users convert after getting an instant answer.

The honest limit: AI is a layer, not the product

Here’s what the hype skips. Khan Academy itself admitted that only about 15 percent of students with access to Khanmigo actually use it, and that practice, not the AI, still drives the measured learning gains. That’s the most useful data point in edtech AI right now.

The lesson isn’t to skip AI. It’s to put it on top of a product that already works. An adaptive tutor on a weak course just helps students leave faster. Fix the core learning loop first, then let AI amplify it.

How to start: the foundation before the features

Clean and tag your content. Lessons mapped by skill and difficulty, in one system. AI personalization is only as good as the map it reads.
Pick the one metric AI can move. Usually completion or retention. Instrument it, then apply AI to that specific step.
Buy the model, build the pedagogy. Use an existing model API. Your edge is your content and your teaching method, not a custom model.
Keep a human on what ships. Generated content and grades get reviewed until you trust the failure rate.

We took the same layered approach across 50 SMB AI rollouts, and sequencing is what separated the ones that stuck from the ones that stalled.

The bottom line

Edtech AI isn’t a bet on the future. Khan Academy, Duolingo, and Coursera already run adaptive learning, AI content generation, and instant tutoring in production. The gap for a small edtech isn’t technology, it’s order: a working learning loop first, clean tagged content next, AI as the amplifier last.

At Kreante we build these implementations for edtech and training businesses, usually on top of the content and tools they already own. If you want to map which of the three fits your product first, book a free AI audit call.