Skip to content

Choose the right path

The most common waste in early AI adoption is building in the wrong place — a throwaway task turned into a full Copilot Studio agent, or a production workflow trapped inside a personal Agent Builder agent. This page is the empowerment team's routing tool: answer a few questions and land on the surface that fits.

Unofficial — a guide, not a rule

This routes the typical case. Real decisions also weigh licensing, data sensitivity, and who will own the result. Use it to start the conversation, then confirm against Licensing & Prerequisites and Microsoft's which Copilot is right for you hub.

The decision tree

Lead with the process, not the product. The first question is simply how often you'll need this — a one-off versus a repeatable need — and the shape of the work does the rest. You don't have to know which tools exist to land in the right place.

Prefer to click through it?

The Path Finder walks you through these same questions one at a time and lands you on a recommendation card — two clicks to an answer.

flowchart TD
    Start([You have a need or a process to support]) --> Q1{How often will<br/>you need this?}

    Q1 -->|Once or occasionally| A{What does the<br/>task look like?}
    A -->|A quick, single step:<br/>draft, summarize, answer| Chat[["Stage 1 · Chat"]]
    A -->|A multi-step job to<br/>hand off and let run| Cowork[["Stage 3 · Cowork"]]
    A -->|A specialized job:<br/>research, analysis, facilitation| FPA[["Stage 2 · First-party agent"]]

    Q1 -->|Again and again| B{Who uses it, and<br/>what does it need?}
    B -->|Just me or my team:<br/>a prompt and a few files, no code| AB[["Stage 4 · Agent Builder"]]
    B -->|The whole org: knowledge,<br/>actions and governance| Studio[["Stage 6 · Copilot Studio"]]
    B -->|Engineered: pro-code, autonomous,<br/>custom models or MCP| Foundry[["Microsoft Foundry"]]

Read it as a table

Prefer words to boxes? Find the row whose need matches yours — frequency first, then shape.

If your need is… …and it looks like Go to
One-off · quick A single step — draft, summarize, rewrite, answer — right now Stage 1 · Chat
One-off · specialized Deep research, data analysis, or facilitation Stage 2 · First-party agents
One-off · multi-step A several-step job you'd rather hand off and let run Stage 3 · Cowork
Recurring · simple The same task again and again; a prompt plus a few files, no code Stage 4 · Agent Builder
Recurring · org-wide Real knowledge sources, actions/connectors, lifecycle and governance Stage 6 · Copilot Studio
Recurring · engineered Pro-code, autonomous or triggered, custom models, MCP at scale Stage 7 · Foundry

What each destination means

  • Stage 1 · Chat — the fastest value. If the need is a single task in the flow of work, you almost never need to build anything.
  • Stage 2 · First-party agents — before you build, check what Microsoft already ships. The best agent is often the one you don't have to make.
  • Stage 3 · Cowork — when a task is multi-step but still a one-off, hand the whole thing off rather than building a reusable agent.
  • Stage 4 · Agent Builder — the right home when the same delegated task keeps recurring and a prompt-plus-files agent solves it. No code, personal or team scope.
  • Stage 6 · Copilot Studio — where agents grow up: real knowledge sources, connectors and actions, publishing, monitoring, and governance for org-wide use. Once here, one more choice shapes cost and capability — which harness powers the agent: the standard harness (rules-based, license-covered in M365 channels), the Copilot chat harness, or the autonomous GitHub Copilot harness (credits for all usage, never license-covered).
  • Stage 7 · Foundry — the pro-code frontier: autonomous and triggered agents, custom models, evaluation, and MCP tools at scale.

When in doubt, climb only one rung

If two destinations feel plausible, pick the simpler one first. It's far cheaper to graduate a working Agent Builder agent into Studio later than to over-build on day one. Each stage is designed to make the next feel obvious — see Start Here.

📚 Learn more.

Was this page helpful?

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