Your best prompts are trapped in one person's head

Look at any team using AI and you'll find the same pattern: one person's output is just better. Their drafts need less editing. Their research is sharper. They get in three prompts what takes everyone else ten. And when you ask how, they can't quite tell you. "I don't know, I just kind of know what to ask it."

That person is the most valuable AI user on the team. Their advantage is also the least likely to spread. Not because they're protective. Most would happily help. It's that the thing making them good was never captured in a form anyone else can pick up. It lives in their head, and heads don't scale.

The tacit-knowledge trap

Some knowledge you can write on a whiteboard. Some you only have in your hands. The economist Michael Polanyi named the second kind tacit knowledge: "we know more than we can tell." The classic example is riding a bike: you can do it perfectly and still not be able to explain the balance to someone who can't.

Great AI use is mostly tacit. The strong user knows which constraints matter, what context to front-load, when the model is bluffing, how to phrase the thing so it doesn't hedge. They've built this by reps, not by reading it anywhere. And tacit knowledge has a brutal property: it doesn't transfer by being nearby. Sitting next to your best prompter doesn't make you a better prompter. You'd have to watch them work, ask why after every move, and rebuild their instincts from scratch, which nobody has time for.

So the expertise stays put. The team's AI ceiling becomes "however good our best person is, applied to only the work our best person personally touches." Everyone else keeps typing the obvious question and getting the average answer.

Why the usual fixes don't work

Teams feel this and reach for the obvious tools. They don't hold.

• "Share your prompts in a doc." The good prompter dumps a few into a shared doc. They arrive with no context, no note on when to use them, and no variables, so they're brittle. The next person pastes one, it half-works for their case, they give up and go back to winging it. The doc rots.
• "Run a training session." An hour where the expert explains their approach. It's genuinely useful and almost none of it sticks, because tacit skill doesn't transfer through a slide. People leave inspired and prompt exactly as they did before.
• "Hire more good prompters." You can't. There aren't enough, and the ones you have got good the slow way. You can't recruit your way out of a knowledge-transfer problem.

Each of these tries to move the skill from one head to another. That's the wrong target. You don't need everyone to become the expert. You need the expert's question to become portable.

Package the question, not the person

Here's the reframe. What your best prompter actually has isn't a talent you have to clone. It's a set of concrete artifacts they carry invisibly: the prompts they'd reach for, the context they'd front-load, the steps they'd run in order. Those artifacts can be extracted, even when the intuition behind them can't be taught.

A good prompt template is a captured expert question. It carries the fields a specialist would have insisted on, the constraints that matter, the checks they'd run, with the specifics pulled out into variables anyone can fill in. A skill goes further: it captures a whole workflow (goal, trigger, steps, output shape) so a junior person runs the senior person's process, not just their words.

When a beginner fills in that template or runs that skill, they don't become an expert. They borrow the expert's articulation: they run a question they couldn't have written themselves and get output they couldn't have produced alone. The tacit trap doesn't get solved by making everyone tacitly skilled. It gets routed around by making the skill explicit once.

What this looks like on a team

The move is to turn your best person's invisible process into shared, runnable artifacts: a prompt library and a set of skills the whole team installs from one place. Then the economics flip:

• The expert writes the process down once, instead of re-explaining it in ten Slack threads.
• Everyone else runs it instead of reinventing a worse version.
• When the expert improves a step, the improvement propagates. The next time anyone runs it, they get the better version. No re-training, no re-sharing.
• The knowledge stops living in one head, so it survives that person taking a week off, or leaving.

That's the point of Korvai for a team: not "store prompts." That's the mechanism. The point is to make the expert's question portable. Hand your team one plugin URL; they install once; your edits show up in their next session. The person who knew what to ask, packaged so the people who didn't can ask it anyway.

Your best prompter isn't hoarding their advantage. It's just stuck in the one place it's hardest to copy from. Get it out of their head and into something runnable, and their ceiling quietly becomes everyone's floor.