Skip to content

title: "One skill, three surfaces: the customer-health capability" description: Follow one skill — ranking accounts worst-first by risk — across three surfaces: a Cowork recipe, an always-on Scout autopilot, and a governed Studio tool. stage: studio roles: [end-user, manager, champion, maker] tags: [skill, lineage, cowork, autopilots, scout, studio, customer-health, three-surfaces] level: intermediate time: 12 min status: walkthrough updated: 2026-08-21

One skill, three surfaces: the customer-health capability

A skill isn't tied to one product. The same reusable capability can be described in Cowork, equipped onto Scout, or hardened into a governed tool in Studio — you pick the surface by how often it runs, how autonomously, and how much governance it needs. This page follows one concrete skill across all three.

Stage: Cross-surface (Cowork → Autopilots → Studio) · For: End user, Manager, Champion, Maker · Level: Intermediate · Time: 12 min

When to use this

Read this when you've captured a reusable skill and need to decide where it should run — on demand in Cowork, always-on in Scout, or hardened into a governed tool in Copilot Studio. It's a decision guide, not a build guide; each surface links out to its own step-by-step walkthrough.

What you'll need

  • A skill you can state in one plain-language sentence (the example below)
  • A rough sense of how often it runs, how autonomously, and how much governance it needs
  • Familiarity with the three surfaces: Cowork, Scout / Autopilots, and Copilot Studio

Try it now

The whole page follows one capability, stated in plain language:

Rank my accounts worst-first by risk, tell me which ones are moving the
wrong way, and don't contact any customer without checking first.

That's a skill — a reusable capability, not a one-off prompt. What changes across the three surfaces isn't what it does; it's who runs it, how often, and under how much control. Same instinct — capture it once — expressed three ways.

Step by step

The same skill, expressed across the three surfaces — read down and stop at the one that matches the job.

Surface 1 — Cowork: the skill you describe and run on demand

On the Cowork surface, the skill lives as a recipe: a multi-step task you capture once — name, inputs, expected output — so anyone can rerun it cold instead of reinventing the prompt.

  • You kick it off when you want it (Monday morning, before a QBR, when a deal wobbles).
  • It runs in the moment, grounded in what Cowork can see, and hands you the ranked list.
  • Nothing is standing or autonomous — it's a saved capability you pull.

→ Build it: Generate a weekly customer-health digest across your accounts

This is the cheapest place to prove the skill

Get the ranking logic and the "what counts as risk" definition right here, on demand, before you ever hand it to an always-on agent. A recipe that's still fuzzy shouldn't be promoted.

Surface 2 — Autopilots (Scout): the same skill, equipped onto an always-on agent

When the recipe is stable and worth running continuously, you graduate it onto Microsoft Scout — the first Autopilot. Now the identical capability is equipped, not described:

  • Scout runs it in the background on its own trigger — the at-risk list re-scores itself as signals change and never goes stale.
  • It works on your behalf under your governed identity, and keeps you in the loop — it surfaces and flags; you decide and it never contacts a customer without your say-so.
  • You stopped starting the skill. You just get pinged when something moves.

→ Equip it: Equip Scout with a continuous customer-health watch

Emerging and gated

Scout is in Frontier private preview — availability, and how you equip and manage its skills, roll out by tenant, license, and region. Treat the Microsoft Scout Learn page as the source of truth before any customer-facing claim.

Surface 3 — Studio: the skill hardened into a governed tool

When the capability needs to take real, governed actions against a system of record, run org-wide under central control, or be shared and monitored like a product, you've outgrown equipping an agent — you build it in Copilot Studio.

Here the terminology sharpens: on the Studio surface a reusable capability becomes a tool (Studio reserves the word skill for a narrower, pro-code-agent meaning — see the Skills Catalog note). The customer-health skill hardens into a small stack of governed tools:

That's the same customer-health skill — now durable, auditable, and shareable, running against real data under central control.

Pick the surface by the job, not the product

The three surfaces aren't a ladder you must climb — they're a choice. Match the surface to what the skill actually needs:

If the skill… Run it on… Because
Runs occasionally, when you decide Cowork (recipe) Cheapest to build and change; you pull it on demand
Should run continuously, on its own, for you Scout (autopilot) Always-on, re-scores itself, keeps you in the loop
Must take governed actions on systems, org-wide Studio (tool) Real actions, policy, monitoring, sharing

Make it better

A skill isn't fixed to one surface. Graduate it as it earns more autonomy and reach — these are the signals it's ready:

Promotion signals — when to graduate a skill

  • Cowork → Scout: you're rerunning the same recipe on a rhythm and wish it just ran itself.
  • Scout → Studio: you need it to act on a system of record, run org-wide, or be governed and monitored — beyond what an equipped agent should do under one person's identity.

Watch out for

  • Don't skip the cheap proof. Promoting a fuzzy recipe straight to a governed Studio tool bakes the fuzz into production. Prove the logic in Cowork first, tighten it on Scout, then harden it.
  • Terminology shifts on the Studio surface. Cowork and Scout call it a skill; Studio building blocks (connector actions, flows, MCP tools, triggers) are tools. A Studio skill is a narrower, pro-code thing — details here.
  • Governance scales with autonomy. The more autonomously and broadly the skill runs, the more the guardrails matter — which is exactly why the customer-health tool ends with governance, not an action.
  • Scout is private preview. Verify behavior against the Microsoft Scout docs if it differs from what's here.

Where this leads

Follow the same capability onward, or build any single surface end-to-end:

Screenshots

This page is a conceptual map, so the visuals live on the three build walkthroughs it points to:

Was this page helpful?

Found an issue or want to suggest an improvement? Open a GitHub issue  ·  Star on GitHub