What Claude Skill Builders Can Learn From OpenAI

It's easy to read the ChatGPT for Teachers launch as bad news if you build on someone else's model, and the last two pieces didn't sugarcoat the squeeze. But there's a second way to read the same launch, and it's more useful. Set aside that OpenAI is a competitor in this instance and look at what it actually did as a piece of go-to-market execution. Because underneath the money and the model, OpenAI just ran a textbook playbook for winning a vertical — and every move in that playbook is available to a solo builder shipping a single Claude Code skill. The lessons don't require a foundation model. They require picking the right narrow thing and distributing it the right way.

This piece is for this site's actual audience: people building AI skills, not people running foundation labs. The claim is simple. The strategy that let OpenAI land in K-12 is the same strategy that separates a defensible skill from a disposable wrapper, and you can copy it deliberately.

Lesson one: go embarrassingly narrow

OpenAI makes a general-purpose model that can do almost anything. And when it decided to win teachers, it did not point the general model at education and call it a day. It built a specific product for a specific user — ChatGPT for Teachers, shaped for the K-12 educator, with the file uploads, connectors, and shareable templates that job actually needs. The most capable generalist on the market chose to go narrow on purpose, because narrow is what wins a vertical. The generality was the asset; the narrowing was the strategy.

The lesson for a skill builder inverts the usual instinct. Most people building on Claude reach for breadth — a "productivity assistant," a "writing helper," a "coding companion" — because breadth feels like a bigger market. It's the opposite. Breadth is where you compete directly with the base model, which is already a broad assistant and always will be. A skill earns its existence by being narrower than the model bothers to be: not "help me write" but a skill that enforces one publication's house style through a specific editorial checklist; not "help me teach" but a skill that runs one specific pedagogical loop and refuses to break it. The base model is the generalist. Your leverage is everything the generalist won't commit to. If OpenAI — sitting on a frontier model — concluded it needed to go vertical to win teachers, the builder with far less leverage should conclude it twice as hard.

The base model is a great generalist and will keep getting better at being general. That's not the ground you fight on. You win on the specificity a general assistant can't afford to have.

Lesson two: distribute through a trusted intermediary

OpenAI didn't try to reach 400,000 teachers one download at a time. It went where teachers already trust an institution. The AFT National Academy for AI Instruction — a $23 million effort by the American Federation of Teachers with the UFT, OpenAI, Microsoft, and Anthropic, aiming to reach 400,000 educators over five years — is a distribution channel dressed as a training program. It puts the tools in front of teachers through the union they already belong to, carried by an intermediary those teachers have reason to trust. That's not incidental to the strategy. For a skeptical, time-poor, institutionally-embedded audience, arriving through the trusted body is worth more than any amount of direct marketing.

The generalizable lesson is that distribution beats discovery, and trusted distribution beats cold distribution. A skill sitting in a repository that nobody knows to look for is a tree falling in an empty forest — it can be excellent and still reach no one. The builder's version of the AFT move is to get your work into the channels your users already trust: a curated marketplace like this one where people go specifically to find purpose-built skills, the community your target users already gather in, the workflow they're already running. You don't need a union. You need to stop assuming that "it's on GitHub" is a distribution strategy, and start treating placement in a trusted channel as a first-class part of the build. OpenAI, which could reach anyone it wanted, still chose to be carried in by an intermediary. The signal there is loud.

Lesson three: free is a habit strategy, not a price

The most misread part of the ChatGPT for Teachers move is the price. Free is not a discount. It's a habit-formation mechanism. The point of giving teachers unlimited access through June 2027 isn't the goodwill — it's that a teacher who spends a school year building their planning workflow around a tool has adopted a habit that doesn't casually reverse. Free removes every barrier to the first use, and the first use, repeated, becomes the default. The vendor is buying habituation, and habituation is far stickier than any contract.

For a builder, the translatable insight isn't "give everything away and hope." It's that the first-use barrier is the real enemy, and lowering it is worth more than capturing value on day one. A skill that's trivial to install and immediately useful on the first run builds the habit that makes it indispensable later; a skill that demands setup, configuration, and a leap of faith before it does anything useful loses most of its potential users at the door — no matter how good it is once you're through it. Design for the first sixty seconds. Make the initial value obvious and unearned. Whatever you eventually charge for — depth, integration, support, the advanced workflow — you earn the right to charge for it only after the habit exists, and the habit only forms if the first use was free of friction.

Where the analogy stops — and why that's the good news

Copy the playbook, not the position. OpenAI's advantages — the model, the capital, the brand — are not available to you, and pretending otherwise leads to building a thin generalist wrapper and losing to the base model, which is the failure mode the earlier pieces in this cluster traced in detail. The vendor can afford to be broad because it owns the model and monetizes distribution. You can't, which is exactly why you shouldn't try to be broad.

But the parts of OpenAI's playbook that actually did the work — going narrow, distributing through trust, removing first-use friction — cost nothing but discipline, and every one of them favors the small, specific builder over the sprawling platform. A single well-made skill that does one real job, lands in a channel its users already trust, and delivers value on the first run is playing the same game OpenAI played to win teachers, just at a scale where narrowness is a superpower rather than a sacrifice. The threat and the lesson arrive in the same announcement. The builders who read only the threat will either freeze or over-react. The ones who read the lesson will notice that the most sophisticated company in the field just demonstrated, at enormous expense, exactly what a defensible AI product looks like — and that the demonstration is free to copy.

Part 94 of 100 in the ChatGPT for Teachers series. Previously: The Squeeze on Third-Party Classroom AI Tools. Next: Where AI Tutoring Skills Fit Between the Giants. Browse more builder insights or explore AI skills for education at aiskill.market.