microsoft/m365-copilot-studio-what-to-build-where

microsoft/m365-copilot-studio-what-to-build-where

Language: TypeScript

License: MIT

Stars: 0

Forks: 0

Open issues: 0

Created: 2026-10-13T05:38:12Z

Pushed: 2026-10-13T06:39:26Z

Default branch: main

Fork: no

Archived: no

README:

Agent platform guidance (Draft)

A guided Astro site for choosing an agent platform from draft decision guidance. It supports platform owners, Center of Excellence (CoE) leads, and platform architects assessing organizational agentic use cases. It explains the fit, alternatives, unmet requirements, and claims that need review. The six outcomes are Agent Builder, Copilot Studio with the GitHub Copilot Harness or Standard Harness, Cowork, Microsoft 365 Agents Toolkit, and Microsoft 365 Agents SDK.

Run locally

Requires Node 22.12+ and npm 9.6.5+.

npm ci
npm run dev

Open http://localhost:4321/. Answers stay in tab memory and reset on refresh. The static comparison page is usable without JavaScript. There are no AI calls, accounts, analytics, answer submission, or remote font requests.

Checks

npm run check
npm test
npx playwright install chromium
npm run test:e2e
BASE_PATH=/m365-copilot-studio-what-to-build-where/ npm run test:e2e

Browser tests build the site and start/stop their own preview server on port 4327. They cover desktop/mobile, themes, the six result candidates, uncertainty, navigation, conditional answers, and no-JavaScript fallback. Unit tests exercise all 15,120 combinations of the encoded answer categories, including all nonempty job and channel combinations, plus partial-answer exclusions, independent control choices, and restoration of platforms after an answer changes.

npm run build
npm run preview

Static output is in dist/. No server adapter is required.

Guidance and rules

The supplied table is draft guidance requiring capability validation. The original table and all review comments are preserved in [docs/source-decision-table.md](docs/source-decision-table.md). Keep this source reference separate from subsequent editorial and implementation changes. Additional source explanations are preserved in [docs/source-question-guidance.md](docs/source-question-guidance.md). Question hints and answer descriptions summarize this material. Informational process guidance shares the knowledge option; processes that write records use the business-actions option. The additional source remains separate from the implemented decision criteria and does not automatically change routing rules. Editorial corrections have been incorporated, but uncertain claims have not been silently resolved. Every visitor sees the draft status. Required unknown capabilities produce a Needs review result rather than an unqualified winner.

• src/data/guidance.ts: question definitions, six products, watch-outs,

and confirmation notes.

• src/lib/candidates.ts: partial-answer capability checks and requirement conflicts.
• src/lib/flow.ts: relevant questions and choices, normalized answers,

and the live shortlist.

• src/lib/recommend.ts: deterministic selection from remaining candidates.
• src/scripts/finder.ts: in-memory answers, navigation, rendering, focus.
• src/pages/compare.astro: public methodology and static reference.
• tests/recommend.test.ts: expected journeys and requirement invariants.

Every answer updates three groups: Still applicable, Requires confirmation, and Excluded. Unanswered questions impose no restrictions. Confirmations stay in the active shortlist; exclusions include reasons and are reconsidered when answers change. Contradictory requirements produce an early no-match review. The sidebar switches between Platform shortlist (the default) and Your requirements using keyboard-accessible tabs. Both views preserve the current answers and update as selections change. Reset answers stays beside question progress on every step and result. It clears all selections and returns to the first question and full shortlist.

Recommendations are selected only from active candidates. Preference precedence is own runtime; voice/handover/existing Standard Harness; owned autonomy; Microsoft 365 API/MCP integration; organization/audience/channel requirements; personal autonomy; Copilot Studio actions/reasoning; knowledge-only work. The first applicable preference among remaining candidates wins. Multiple remaining options without a distinguishing preference produce a review result. Excluded platforms remain in the shortlist's exclusion section, outside the result's alternatives.

The first question accepts multiple jobs: knowledge or process guidance, business-system actions, and complex tasks/documents. Every selected job contributes its requirements; selecting knowledge alongside actions or reasoning does not restore a platform excluded by those needs. The summary lists all selected jobs.

The second question always asks who guides the work, as a separate single choice. Job selections never infer ownership. Employee-directed business actions retain Cowork subject to plugin confirmation; user direction alone does not select it. Editing the selected jobs preserves the explicit control answer.

Audience is the third question. The Microsoft 365 knowledge/guidance job is explicitly employee-facing: selecting it, alone or with other jobs, hides both customer audience choices. The audience step remains visible with Employees preselected and an explanation, so visitors can continue or go back. Continue records the selection; going back before confirming leaves it unanswered. Adding that job clears an incompatible audience answer and requires confirmation; removing it restores the customer choices. This is the scope of the job option, rather than a universal restriction on agents that use Microsoft 365 knowledge sources.

Multi-channel requirements follow from selecting more than one of Microsoft 365, website/custom app, and voice. Live-human handover is a service capability and does not count as an additional channel.

Follow-up visibility is evaluated from earlier answers and the candidates still active at that point. API/MCP integration questions disappear once Toolkit is excluded. The existing-Standard-Harness question is skipped when voice/handover already determines the harness preference, or the Copilot Studio options are excluded. Questions about audience, channels, and autonomy still capture prerequisites when SDK is the sole remaining candidate. Hidden...