Plan and configure agent solutions
Topics, Responses, and Conversation State
CoreUse topics, tools, prompts, knowledge, HTTP calls, generative answers, adaptive cards, and variables to control conversation behavior.
Aligned to the AB-620 live Skills measured outline; visible page date April 21, 2026; product behavior verified September 19, 2026.
Why this matters
Topics provide explicit conversational control around generative orchestration. The strongest designs know when to preserve deterministic branching and when to let grounding or tools produce a response.
Must Know
- Add agent flows to a topic: Call the agent flow from the topic, map topic values to flow inputs, and use its outputs in subsequent branches and messages.
- Configure agent response formatting: Use explicit response instructions or structured output so the answer matches the required channel, schema, tone, and content boundary.
- Add tools to a topic: Invoke a specific tool in the topic when the conversation path requires deterministic control and map its inputs and outputs.
- Configure advanced agent responses with custom prompts: Create a custom prompt with a bounded task, grounded inputs, model settings, and a defined output that later nodes can use.
- Configure advanced agent responses with custom knowledge sources: Supply the intended knowledge source and query context to ground the response while preserving source permissions and citations.
- Configure advanced agent responses with API and Send HTTP requests: Configure the endpoint, method, headers, authentication, request body, response parsing, timeout, and safe error branches for the HTTP call.
Compare and Distinguish
- Generative orchestration selects topics, tools, and knowledge from descriptions and context; classic orchestration primarily routes through recognized trigger phrases.
- A generative answers node grounds a generated response; a custom prompt performs a specified generative task; an HTTP request calls an endpoint directly.
- Topic variables are local to a topic; global variables persist across topics in the same conversation; system variables are supplied by the runtime.
Scenario examples
- Scenario: A pilot succeeds but production users have different permissions. Think: trace identity and authorization through every source, tool, and delegated agent.
- Scenario: A component works in development but fails after deployment. Think: inspect solution dependencies, connections, environment configuration, and channel-specific behavior.
- Scenario: A response is fluent but the operational result is uncertain. Think: separate generation from retrieval, tool execution, telemetry, and verification.
Exam traps
- Using a generative answer where a regulated transaction requires a deterministic action and validation.
- Passing secrets in a prompt, variable, card, or raw HTTP body that can appear in logs or transcripts.
- Assuming an adaptive card behaves identically in every publishing channel without channel testing.
Key takeaways
- Choose each authoring element by responsibility: topics sequence conversation, tools and flows perform operations, knowledge grounds answers, prompts generate bounded content, and variables carry state.
- Make the identity, data, lifecycle, and failure boundaries explicit for configure topics.
- Validate the behavior with representative evidence before broad deployment.
How it works
- Copilot Studio uses the configured descriptions, instructions, conversation state, and orchestration mode to select the relevant topic, knowledge, tool, flow, or connected agent.
- Runtime identity and environment configuration determine whether the selected capability can access its dependency and return a governed result.
Objects and administrative surfaces
- Topic authoring canvas for nodes, conditions, variables, cards, tools, and generative answers.
- Tools and Knowledge pages for reusable capabilities and grounding sources.
- Test pane activity map for observing topic, tool, and knowledge selection.
When to use it
- Use this topic when the requirement maps to configure topics.
- Prefer the narrowest supported surface that satisfies the required behavior and governance boundary.
Security and governance implications
- Apply least privilege to maker, runtime, connector, source, and deployment identities.
- Classify data and connectors, preserve audit evidence, and avoid exposing secrets or sensitive payloads in prompts and telemetry.
Troubleshooting signals
- Reproduce with the same version, channel, identity, inputs, and environment before changing configuration.
- Use activity traces, flow runs, tool errors, evaluation results, or telemetry to locate the failing layer.
More detail
- With generative orchestration, the current conversation and capability descriptions drive which knowledge source, topic, or tool is chosen; author a topic when the experience must enforce a specific sequence, branch, or invocation.
- When a topic calls a flow or tool, map topic values into the capability’s inputs and retain its structured outputs for later conditions, messages, or error branches.
- Use a custom prompt for a bounded generative task with defined inputs and outputs; use generative answers when the requirement is to retrieve evidence and ground an answer in configured knowledge.
- For knowledge-backed responses, pass the intended query context, preserve the source’s permissions, and make citations part of the evidence that the answer is grounded.
- Treat Send HTTP request as a direct integration contract: build the call from its method and endpoint, apply authentication and headers, serialize request data, parse the response, and route timeout or failure outcomes safely.
- Match response formatting and adaptive-card structure to the target channel, and choose topic, global, or system variables according to how long and where the conversation state must remain available.
Ready for the quiz?
- Which requirement determines the correct surface for configure topics?
- Which identity performs the operation, and where is authorization enforced?
- What evidence distinguishes a safe successful result from a fluent but incomplete one?
Related objectives
- D1.3.S1 — Add agent flows to a topic
- D1.3.S2 — Configure agent response formatting
- D1.3.S3 — Add tools to a topic
- D1.3.S4 — Configure advanced agent responses with custom prompts
- D1.3.S5 — Configure advanced agent responses with custom knowledge sources
- D1.3.S6 — Configure advanced agent responses with API and Send HTTP requests
- D1.3.S7 — Configure generative answers node
- D1.3.S8 — Configure adaptive cards
- D1.3.S9 — Manage variables