Copilot + Power Automate: Handling long‑running HTTP calls without Copilot appearing “stuck”

(3) Share

Report

Posted on by RichAI

Hi everyone,

I’m working with Copilot and invoking Power Automate flows from Copilot to call external HTTP APIs.

A pattern I’m running into is:
Copilot invokes a flow
The flow calls one or more HTTP actions
While the flow is running, Copilot provides no intermediate feedback

From the user’s perspective, Copilot appears “stuck” until the flow completes or fails

I understand that today Copilot tool invocation is synchronous and doesn’t receive streaming or intermediate status events from Power Automate runs.
What I’m looking for is:
1. How are others designing around this limitation in real solutions?
2. Are you splitting flows into “start work” vs “complete work” patterns?
3. Are there any UX or architectural patterns that have worked well in regulated or enterprise environments?

I’m looking for supported workarounds, recommended, platform‑aligned patterns that keep Copilot responsive and user‑friendly.
Any examples or lessons learned would be really appreciated.

Thanks in advance!

Categories:

Calling actions from Copilot Studio

I have the same question (0)

All responses (5)

Answers (0)

Sort by

Suggested answer

ninihen 119 on at

Like (4)

Report
Copy link

Link copied!

My 2 cents. I would split this into a start/status pattern rather than making the agent wait for the whole job.

1. Parent flow (Skills-triggered, called by agent) does only the cheap work + returns a Response immediately, e.g. "Started - run ID X."
2. Child flow (HTTP-triggered) does the long work async. Parent fires it via the trigger URL and doesn't wait.
3. The agent has a second "check status" action that reads the job record, flow run, or produced artifact.

My handling with Copilot agent + power automate is using the Flow Studio MCP for Power Automate (disclaimer - I'm one of the people who build it). This tool gives the agent full visibility of what's going on so the agent doesn't rely on the response to know if the flow is finished and agent is not bounded to trigger the flow with the standard agent trigger - agent can trigger generic http flows and resubmit flow runs.

Setup: https://learn.flowstudio.app/blog/copilot-studio-mcp-debug-power-automate

Another pattern is to move the conversation out of Copilot Studio. Use Teams adaptive cards or approvals for the human interaction, and let Power Automate orchestrate the work and call the agent when needed using flow action "execute agent and wait". Here is an example: https://learn.flowstudio.app/blog/power-automate-flow-fixes-other-flows-human-in-the-loop-agent

Was this reply helpful? Yes No
Vish WR 2,244 on at

Like (1)

Report
Copy link

Link copied!

@RichAI

agree with @ninihen

Was this reply helpful? Yes No
Suggested answer

11manish 1,922 on at

Like (2)

Report
Copy link

Link copied!
For your scenario (Copilot + HTTP APIs):

Use:

Start Flow + Async Processing + Status Polling

This gives:

Fast response in Copilot

Better user experience

Clean architecture

Enterprise compliance

Don’t make Copilot wait—make it orchestrate asynchronous work

Was this reply helpful? Yes No
Suggested answer

Valantis 4,803 on at

Like (1)

Report
Copy link

Link copied!

Hi @RichAI,

Good question and you've diagnosed the platform correctly.

The hard limits to design around:

- Copilot Studio enforces a 100-second limit on synchronous flow responses (the Respond to the agent action must fire within 100 seconds)
- There's also a ~2-minute total flow timeout in some configurations
- There is no streaming or intermediate status mechanism from flow to agent

Pattern 1: Split the flow at the Respond to the agent action (confirmed supported)

This is the Microsoft-recommended approach. Place the Respond to the agent action early in the flow immediately after you've done enough work to give the user a meaningful acknowledgment and put the heavy HTTP work after it. Actions placed after Respond to the agent continue running up to the full 30-day flow duration limit without blocking the agent response.

So the flow responds to the agent: "Your request is being processed, I'll notify you when it's done" and then continues running the long HTTP call in the background.

Pattern 2: Proactive message when complete (confirmed supported, Teams only)

After the background work finishes, the flow sends a proactive message to the user via the Teams connector with Post as: Microsoft Copilot Studio agent. This delivers the result directly to the user's personal chat with the agent without requiring a new user message.

Limitations to know: proactive messages only work in Teams, not in web chat or other channels. The user must have the agent installed and not blocked it.

Pattern 3: Express mode (preview)

Microsoft recently released express mode for agent flows which speeds up execution significantly, increasing the likelihood of completing within the 2-minute window. Worth enabling if your environment supports it no extra cost, toggle it on in the flow's Overview page.

For regulated/enterprise environments, Pattern 1 + 2 combined is the cleanest architecture:

1. Agent calls flow
2. Flow validates input, kicks off the HTTP call, immediately responds to agent: "Processing your request"
3. HTTP call completes in the background
4. Flow sends proactive Teams message with the result
5. Result optionally stored in Dataverse for the agent to retrieve on follow-up query

Ref: https://learn.microsoft.com/en-us/microsoft-copilot-studio/flow-modify-use-with-agent

Ref: https://learn.microsoft.com/en-us/microsoft-copilot-studio/advanced-proactive-message

Ref: https://learn.microsoft.com/en-us/microsoft-copilot-studio/agent-flow-express-mode

Best regards,

Valantis

✅ If this helped solve your issue, please Accept as Solution so others can find it quickly.

❤️ If it didn’t fully solve it but was still useful, please click “Yes” on “Was this reply helpful?” or leave a Like :).

🏷️ For follow-ups @Valantis.

📝 https://valantisond365.com/

💼 LinkedIn

▶️ YouTube

Was this reply helpful? Yes No
Suggested answer

Haque 2,323 on at

Like (3)

Report
Copy link

Link copied!
Hi @RichAI,

Missing intermediate feedback while a Power Automate flow is running-Copilot seems “stuck” until the flow completes/fails—is a known user experience challenge when invoking long-running flows from Copilot Studio.

Why this happens: By this time we all know it but let's revise

Copilot Studio waits synchronously a response.

No built-in mechanism to provide progress updates.

No built-in mechanism to provide intermediate status messages during flow execution

Most of the important things Valantis and Ninihen have covered, I can offer some cheaper options :) if you like

Provide Immediate Acknowledgment: Design the agent to immediately respond with a message like “Processing your request, this may take a moment…” before invoking the flow. This sets user expectations and reduces perceived “stuck” time.

Implement Status Polling or Callbacks: Use a status tracking mechanism (e.g., store flow progress in Dataverse or SharePoint). Copilot can query this status on user request or via proactive notifications when the flow finishes. This option is very convenient if have SharePoint alongside.

Optimize Flow Performance: If there are rooms and spaces, minimize the number and complexity of HTTP calls inside the flow. Use parallelism (if possible) and efficient error handling to reduce overall runtime.

User Interface Design: f embedding Copilot in a UI, consider adding loading spinners or progress indicators during flow execution.

I am sure some clues I tried to give. If these clues help to resolve the issue brought you by here, please don't forget to check the box Does this answer your question? At the same time, I am pretty sure you have liked the response!

Was this reply helpful? Yes No