An AI tutor is the feature you add last, not the product you build first

Every pitch for a course product opens on the same slide: an AI tutor that answers a learner's questions at two in the morning, patient and infinite, the teacher who never sleeps. It's a good slide. It gets the nods. It is also the part of this product you should build last, because it's the feature, not the company — and building it first is the most common way this idea turns into an impressive demo that never becomes a business.

The reframe is the whole article, so here it is plainly. The thing you're excited about — the tutor — is rented intelligence wrapped around your content, and it's swappable the day a better model appears. The thing you're not excited about — the course platform underneath it — is the product, the asset, and the long, unglamorous build. The tutor is a feature of the platform. The platform is not a feature of anything. Get that backwards and you build the small part first and discover the large part was the point.

The part nobody puts on the slide

Look at what a course platform actually has to do before a tutor has anywhere to stand. Someone enrolls, and enrollment is a relationship that has to be remembered: who paid, what they bought access to, when it expires, what happens when it lapses. They work through material, and the product has to hold their progress — what they've completed, where they stopped, what they've passed, what unlocks next — because a course product that forgets where you were is one a learner abandons on the second visit. There's the content itself: lessons, modules, the order they're meant to be taken in, the quizzes and the completion that turns "watched some videos" into "finished the course." There's the instructor's entire other half of the product — authoring, pricing, seeing how a cohort is doing. And there's the money, the subscriptions or the one-time purchases, with a record you keep yourself.

That is a large product. It's worth being honest about the size, because this is one of the build ideas in this set where the usual line — a week, if the foundation already exists — does not hold. A foundation carries the payments, the accounts, the content storage, and the jobs, and that genuinely removes months of the work. It does not make a learning-management system small. The enrollment-and-progress spine is the product, it's intricate, and even with everything around it already built, it's a long build. The tutor, by contrast, is a few weeks of work bolted onto the end — which is exactly the inversion worth seeing, because the few weeks are the part that demos and the long build is the part that sells.

Where the tutor actually helps — and why it's a feature

The tutor earns its place in exactly one way: by being grounded in this course's content. A tutor that answers from the open internet is a generic chatbot someone pasted into a sidebar; a tutor that answers from the lesson the learner is currently stuck on, in the vocabulary the course uses, refusing to wander off into material the course hasn't taught yet, is a real feature. That grounding is not a model capability you buy — it's the same discipline the support-chatbot build idea is built around: retrieve from your own material, answer from what you retrieved, and say I don't know rather than inventing. The tutor is rented generation sitting on top of an index built from the course content you own, and the content and the index are the assets. The model is the part you could swap next quarter without a learner noticing.

So the tutor is a feature for the same reason the segmentation logic is the product elsewhere in this set: it's defined by the data it sits on, and here that data — the curriculum, the learner's progress, the owned content — belongs to the platform. Built on its own it's a parlour trick; built on top of the content and the record of where each learner is, it's genuinely useful. The platform makes the tutor possible; the tutor does not make the platform.

What it rides, and what it owns

The chain underneath is mostly foundation furniture: accounts, subscriptions and payments with a local record of who bought what, a content system for the lessons, scheduled jobs for a cohort's drip and its end-of-course certificate, the admin and the log. None of that is the differentiator. What the product owns is the learner's history and the curriculum — the enrollment, the progress, the completions, the structure of the material and the order it teaches in. That record is the asset: what a learner loses if they leave, what an instructor's business is built on, and what the tutor grounds itself in. Own it, and the video host and the model are inputs you plug in; own only a clever tutor, and you have a feature looking for a product to belong to.

The hard part

Three, and the model is the least of them. The first is content — a course platform is only as good as what's taught on it, and that's licensing, authoring, and quality that no API supplies. The second is the build size itself, named above: this is a long road even with the foundation in place, and the discipline is to spend that road on the platform and not be lured into polishing the tutor while the enrollment-and-progress spine is still half-built. The third is the market, which is real and mature — Teachable, Thinkific, Kajabi and others have built the platform half well, for years. The tutor is, at the moment, the freshest reason a new entrant has to exist; but a tutor on top of a weak platform loses to a strong platform that adds a tutor, because the platform is the thing that's hard to catch up on and the tutor is the thing anyone can add in a few weeks.

The verdict

This is a full SaaS, and a substantial one — not a deflating "it's only a feature," but a warning about which part is the feature. Build the course platform first: the enrollment, the progress, the content, the payments, the instructor's side. Add the tutor last, grounded in the content and the progress you now own, where it's genuinely good because it's standing on something. Build it the exciting way round — the tutor first, the platform someday — and you'll have the best demo at the conference and no product behind it. The accounts, billing, content, and jobs the platform rides on are meant to already exist precisely so the long build goes into the learning system that is the company, and the tutor arrives as what it always was: the feature on top.