Approved Expertise: Knowledge Your AI Can Run

Approved expertise — an executable skill — captures how a piece of work actually gets done, precisely enough that any teammate's AI Wingman can act on it. It's approved by the teammate who owns it, and cited to them every time it's used.

It is the unit the Company Brain is made of. Not pages, not transcripts, not embeddings of old Slack threads — the working knowledge itself, with a person's name on it.

What's inside one

Take "Conversion Metric — taught by Mark, Growth." Behind that one line sits everything a teammate — human or AI — needs to get Mark's answer without Mark:

• The definition — paying subscribers ÷ active free users, trailing 30 days. Not "conversion" in the abstract; conversion as this company measures it.
• The data — which tables, which filters, which exclusions.
• The thresholds and exceptions — what counts, what doesn't, and the edge cases Mark learned the hard way.
• The judgment calls — when to distrust the number, what to check next.
• The output — what a finished answer looks like, so quality is reproducible.

Why documentation can't do this job

A document is written for a human who fills the gaps. Approved expertise leaves no gaps to fill.

Ask an AI to "do the pricing review the way we do it" off a wiki page and it improvises the gaps — silently. Approved expertise removes the improvisation: the how is explicit, so a Wingman acting on it produces the author's quality of work, and says so. That's the difference between AI that demos well and AI you can trust.

The lifecycle: solve → approve → acquire → improve

1. Solve. Expertise is born inside real work — a team member and their Wingman get something done that works. Nobody sits down to "write a skill."
2. Approve. The Wingman asks one question: "Want me to add this to the Company Brain as approved expertise?" The human decides. Nothing enters without them.
3. Acquire. Every teammate's Wingman has it instantly — cited to the author. Emma approves her paid-conversion solve, and the next growth question anywhere in the company starts from her playbook.
4. Improve. Constant use surfaces drift fast, and every solve that builds on a piece of expertise sharpens it. The author stays its owner — updating it updates every Wingman at once.

Citation is structural

Every use of approved expertise names its author: "Conversion Metric — taught by Mark, Growth." That does three jobs at once:

• Trust — you know exactly whose judgment an answer stands on, and answers built on approved expertise carry the High confidence mark.
• Credit — teaching the brain is visible, attributed work. People share what they know when their name compounds with it.
• A path to the human — when the question outgrows the expertise, you know who to ask. The brain routes you to Mark; it doesn't replace him.

What approved expertise looks like in practice

One solve drew on the first three and produced the fourth — that compounding is the whole point. See it happen in How Catalyst works.

Frequently asked questions

What is approved expertise in one sentence?

How a piece of work actually gets done, captured precisely enough for any teammate's Wingman to act on — approved by its author and cited to them on every use.

How does something become approved?

By working: solve with your Wingman, approve when it asks. Every teammate's Wingman has it from that moment.

Is this the same as a prompt library?

No. A prompt is a way of asking; approved expertise is the how of the work itself — data, definitions, thresholds, judgment, output — with an owner and an approval behind it.

Who gets the credit?

The author, every time, by name. Attribution is built into the structure of the Company Brain.

What about stale expertise?

Constant use surfaces drift quickly; the author owns updates, and every new solve refines what it builds on.