Skill sharing: bottle what works, deploy it everywhere

In every organisation we work with, AI productivity follows the same pattern: a handful of people get extraordinary results, and everyone else gets mediocre ones. The difference isn’t talent — it’s technique. And technique can be shared.

The problem with prompt folklore

Today, AI know-how spreads by folklore. Someone crafts a brilliant workflow for writing customer proposals, shares it in Slack, and it scrolls away forever. Three months later, four people maintain four slightly different broken copies. The company keeps re-learning the same lessons.

Skills are know-how with a deploy button

Harriet’s skill sharing turns a working technique into a managed artifact:

• Capture it once. A skill packages the prompt, the steps, and the context that makes a workflow reliable.
• Publish it to teams. Roll a skill out to the people who need it — sales gets the proposal writer, support gets the escalation summariser.
• Improve it centrally. When the skill gets better, everyone’s copy gets better. No more folklore drift.

Because skills run on the same control plane as everything else, they inherit your guardrails automatically: the models each team is allowed to use, the budgets they run under, and the regions their data stays in.

Compounding, not copying

The teams that win with AI won’t be the ones with the best individual prompters. They’ll be the ones whose collective know-how compounds — where every discovery becomes part of the company’s permanent capability. That’s what skill sharing is for.

It’s also why we rebuilt Harriet around the control plane: shared capability only works when there’s a layer to share it on. Want to see skills in action? Book a call.

Common questions

What is an AI “skill”?

A skill packages a working technique — the prompt, the steps, and the context that make a workflow reliable — into a managed artifact you can publish to the teams that need it.

How is skill sharing different from pasting prompts in Slack?

Shared prompts drift into stale, divergent copies. A skill is maintained centrally: improve it once and everyone’s version improves, so know-how compounds instead of decaying.

Do shared skills respect our governance rules?

Yes. Skills run on the same control plane as everything else, so they inherit each team’s allowed models, budgets, and data-region policy automatically.