What it does
Tools are how an agent does things rather than just talk: call a Power Automate flow, hit a connector, run an AI prompt, or invoke a REST API. Under generative orchestration, the model decides when to call each one.
Key facts
- Tool types include Power Automate flows, prebuilt and custom connectors, AI Builder prompts, and topics exposed as tools.
- Inputs and outputs are typed; the model fills parameters from the conversation when orchestration is on.
- Connectors run under a connection reference, so authentication and licensing follow the connector, not the agent.
- Tool names and descriptions drive selection under orchestration — write them for the model, not for you.
When to use / skip
Use tools whenever the agent must read or write real systems — create a ticket, look up an order, update a record. Skip building a custom API tool when a certified connector or existing flow already does the job.
Configuration decisions
- Flow vs connector vs custom API for each action.
- Which connections and service accounts back the tools, and their permissions.
- Input validation and confirmation steps before write actions run.
Gotchas
- Premium connectors carry their own licensing — a "free" agent can quietly need premium entitlements.
- The model can call a write action with the wrong parameters. Confirm before anything irreversible.
Consultant notes
- Governance lives with the connectors: DLP policies apply to what agents can touch, so plan them together.
- Reuse existing flows and connection references rather than rebuilding integration logic inside the agent.
- Watch capacity and connector licensing together — both scale with usage and both surprise people.
Recheck available tool types each release.