Why is Copilot Studio Test Session so slow?

(2) Share

Report

Posted on by AW-09010751-0

Testing an agent using the Test Session in Copilot Studio is responding noticeably slower than normal behavior deployed to Teams.

My agent is emitting a Markdown table, and it is like watching paint dry. This makes for a terrible DX.

Is this a setting that I can adjust?

Categories:

General topics

I have the same question (0)

All responses (7)

Answers (0)

Sort by

Suggested answer

Valantis 6,202 on at

Like (1)

Report
Copy link

Link copied!

Hi @AW-09010751-0,

The test pane runs in debug mode. It's drawing the execution path in real-time and capturing variable snapshots for every single step. When your agent outputs something complex like a Markdown table, the browser has to render both the chat output and map all that data to the visual graph. That's what causes the delay. Teams is faster because it just shows the chat without all the debugging stuff.

Possible fix: Open the test pane, click the three dots menu at the top, and turn off "Track between topics" (or "Track between topics"). This kills the real-time visualization and speeds everything up.

Let me know if that worked out for you.

Was this reply helpful? Yes No
AW-09010751-0 32 on at

Like (1)

Report
Copy link

Link copied!

Thanks for the tip, but Track between topics was already off.

Was this reply helpful? Yes No
Suggested answer

Valantis 6,202 on at

Like (1)

Report
Copy link

Link copied!

Hi @AW-09010751-0,

A couple other things to try is republish your agent, then log out and log back in. Sometimes testing in a different browser session helps too.

Let me know if that worked out for you.

Was this reply helpful? Yes No
crob 60 on at

Like (1)

Report
Copy link

Link copied!

My test sessions are not that slow. They are not fast either. But editing the adaptive card JSON is REALLY slow, the max I can type is 1 character per second, sometimes less.

I'm in a non-chargeable environment, maybe that's related.

Was this reply helpful? Yes No
Valantis 6,202 on at

Like (1)

Report
Copy link

Link copied!

Hi @AW-09010751-0,

This is all we can check. There’s nothing documented by Microsoft about slow adaptive cards. I’m on the paid version and it works fine and fast for me. Replies sometimes take a moment, especially when I use Topics and flows.

Maybe try Incognito/InPrivate or a different browser (and delete cookies/cache)that might help.

I hope this helps. If you found the info so far useful, please mark the answer as “Answered” so others with similar issues can find it more easily.

Was this reply helpful? Yes No
Suggested answer

Valantis 6,202 on at

Like (1)

Report
Copy link

Link copied!

Hi @AW-09010751-0,

Just wanted to check in and see if everything is working now. If you still need any help, feel free to let me know.

Also, if the issue is resolved, it would be great if you could mark the answer as solved so others with the same question can find it easily.

Thanks and have a great day!

Was this reply helpful? Yes No
AW-09010751-0 32 on at

Like (0)

Report
Copy link

Link copied!

Thanks for following up @Valantis, but my experience is still that everything is super slow with Copilot Studio.

A big difference between my first attempts at building AI Agents and later work was that I was relying heavily on Agent Instructions. I basically gave up on that approach because it was super unreliable. It would frequently skip steps and randomly hallucinate. I had to shift to rely more on Agent Flows. Agent Flows generally respond faster (and are easier to debug) and I often use them to construct Markdown, which tends to emit faster.

However, the overall DX is incredibly poor:

- It usually takes several seconds to bring up a screen or switch tabs.

- Debugging a Topic is almost impossible -- I know which nodes are hit, but can't interrogate the variables.

- Painting the Topic workflow is slow and janky.

- Clicking on buttons (like a formula) takes seconds to react.
- Many input boxes need content, but are micro small.
- The expression languages (between Topics and Agent Flows) are inconsistent and both lacking in functionality.
- I can't trade JSON structures between Topics and Agent Flows without first doing a String conversion.
- No native git synchronization (or poor version history with no concept of what changed or a diff).

Overall, the interface is not responsive enough. And I don't even have a huge project yet -- only 4 AI Agents, 12 Agent Flows, and 5 Prompts. Imagine how much worse it will be if our project was larger? I can't imagine this tool scaling well with no folders, limited reuse (can't really re-use a Topic without making copies), and no git support.

I'm hoping Microsoft is making efforts to speed things up and address product limitations.

Was this reply helpful? Yes No