Where AI Tutoring Skills Fit Between the Giants

Look closely at what the two biggest AI labs actually built for education and you'll notice they built the same thing. ChatGPT for Teachers and its counterpart from Anthropic are both, at their core, teacher-facing productivity tools: they help the adult in the room plan faster, differentiate materials, draft communications, and generally get the paperwork of teaching done in less time. That's a genuinely valuable job. It is also, conspicuously, not the job of teaching a student. There is a whole category of software the giants have circled and not entered — the student-facing tutor — and the gap between what they shipped and what a learner actually needs is where the durable skill-builder opportunity lives.

This is the payoff piece of the builder cluster, and it connects the whole ChatGPT for Teachers story back to something this site has argued at length in its AI Tutoring series. The argument there and here is the same: good tutoring is a specific, encodable behavior — a skill — and it's precisely the behavior the general-purpose teacher tools aren't built to perform.

Both giants aimed at the teacher, not the student

Notice the shape of the products. ChatGPT for Teachers is a workspace for the educator — uploads, connectors, shareable templates, all oriented around helping a teacher produce and organize. The paid platforms in the same market, surveyed in the enterprise comparisons, sit in roughly the same posture: tools an institution buys so its staff can work faster. The user is the adult. The output is materials. The job is productivity.

This is not an oversight, and it's worth understanding why both giants landed in the same place. Teacher productivity is a safer, broader, more monetizable target than student tutoring. Adults can evaluate the output and catch the errors. The safety surface is smaller — you're helping a professional make a worksheet, not sitting unsupervised with a twelve-year-old. And "makes teachers faster" is an easy institutional sell. Student-facing tutoring is the harder, riskier, narrower job: it needs guardrails a productivity tool never has to think about, it has to withhold help rather than maximize it, and it has to be right in ways a draft-a-parent-email tool doesn't. Rational companies serving the broad market chose the broad, safe target. That choice is exactly what leaves the narrow one open.

The giants built tools that help the adult make materials. Neither built the tool that sits with the learner and, at the moment they're stuck, refuses to just hand over the answer.

The tutoring job the productivity tools can't do

A teacher-productivity assistant and a student tutor are optimized for opposite behaviors, and this is the crux of the whole argument. A productivity tool is measured by how much it hands you — the faster and more complete the output, the better it's working. Ask it for a lesson plan and it should produce a great lesson plan, immediately, in full. Maximal helpfulness is the entire point.

A tutor measured that way is a disaster. As this site's tutoring series argues, the core tutoring move is withholding — asking a question when it would be faster to tell, giving the smallest scaffold that lets the learner take the next step themselves, refusing to hand over the answer at the first sign of struggle because struggle is where the learning actually happens. A general assistant tuned to be maximally helpful does the opposite by default: it solves the problem beautifully and completely, the student copies it down and understands nothing, and comes back tomorrow just as stuck. The behavior that makes a productivity tool good is the behavior that makes a tutor useless. You cannot get one by lightly reskinning the other, because they're pointed in opposite directions.

That opposition is the whole opportunity. The pedagogical restraint a real tutor needs — diagnose before instructing, question before telling, scaffold in sized steps, refuse the answer, check that it landed — is not something the giants' teacher tools do, because doing it would make them worse at the job they're actually built for. It has to be installed as a specific behavior, on top of a capable model, by someone who cares about the pedagogy enough to encode it. That someone is a skill builder, and the thing they build is a purpose-built AI tutoring skill.

Why a skill fits the gap the giants leave

The reason this is a skill-shaped opportunity and not a foundation-model one is that the missing piece was never intelligence. The models are already more than capable enough to tutor well; they simply aren't instructed to. What's missing is the governance — the file that tells the agent when to question, when to scaffold, when to hold the line and when to give a hint — and that governance is exactly what a skill is. Subject-specific pedagogy compounds the point: a Socratic maths tutor that walks a student through their own reasoning, a science tutor that runs predict-observe-explain so the learner guesses before seeing the result, an English tutor that refuses to rewrite the student's paragraph because their voice is the point. Each is a distinct, encodable procedure. None of them is something a broad teacher-productivity workspace is trying to be.

This is where the whole series converges. The free vertical product from the model vendor flattens the broad middle of EdTech — the generic lesson-planning wrappers — and the right builder response is to go narrower than the giants ever will. Student-facing Socratic tutoring is that narrower place, drawn with unusual clarity: it's a job both giants deliberately stepped around, it demands a behavior their products are structurally built not to have, and it's expressible as a specific skill rather than requiring a frontier model of your own. The moat isn't capability. It's the pedagogy you're willing to encode and defend.

The map, plainly

Here is the landscape the giants have drawn, and the opening they left. ChatGPT for Teachers and Claude for Teachers own teacher productivity — planning, differentiation, communication — and they own it well, one of them for free. That ground is taken; don't build a generic version of it. The paid enterprise platforms own institutional integration and admin controls. Also largely taken. What remains unclaimed, in plain sight, is the student-facing tutor: the tool that sits with a learner, runs a real pedagogical loop, and withholds the answer on purpose. Neither giant is building it, because it's narrow, hard, and pointed against the grain of a productivity tool.

For anyone building AI skills, that's not a gap to lament — it's a map. The giants have helpfully marked the territory they'll keep absorbing and, by omission, the territory they'll leave alone. Purpose-built tutoring skills fit exactly in the space between them: too narrow and too pedagogically specific for a company chasing every teacher at once, and too behaviorally particular to fall out of a general model without someone deliberately installing it. That's the whole thesis of this site's AI Tutoring series, and the ChatGPT for Teachers launch, read carefully, is the strongest evidence yet that the thesis is right. The giants didn't take the tutoring job. They confirmed it was still there for the taking.

Part 95 of 100 in the ChatGPT for Teachers series. Previously: What Claude Skill Builders Can Learn From OpenAI. Next: Building the Skill a Free Chatbot Can't Replace. Browse more builder insights or explore AI skills for education at aiskill.market.