AI Catalog/Agent Usability/Docs problem

Description

(copied from initial report in internal slack):

AGENT USABILITY ISSUE REPORT:

So, in response to @Darva Satcher’s mission and requests to get us all using more Duo and Agents...

I put on my "Dumb User" hat (I wear that one all the time), and tried to figure out how I actually am supposed to USE these agents from the agent catalog.

I started from a link to an agent in the agent repo (similar to one that was shared with me yesterday in the Duo Agentic Platform Office hours): https://gitlab.com/explore/ai-catalog/agents/105/ , which is Epic/Issue summarizer.

But I didn't know how to USE that agent. I'm sure there's docs "somewhere" (or I could have watched Darva's video), but I have my dumb user hat on and want to use Duo, so I went to an issue page, and opened Duo Chat panel to ask it to explain it to me.

The resulting conversation was sort of helpful, but also very confusing and misleading in other ways, and left me with more questions than I started with.

Session ID was 1234970. I'll include screenshots of the session in thread.

TL;DR: I think we could do much better with the help/guidance/UX on how to actually use agents from the agent catalog.

Screenshots

Screenshot_2025-11-04_at_11.05.22_AM Screenshot_2025-11-04_at_11.10.08_AMScreenshot_2025-11-04_at_11.10.31_AMScreenshot_2025-11-04_at_11.10.48_AMScreenshot_2025-11-04_at_11.10.57_AMScreenshot_2025-11-04_at_11.11.08_AMScreenshot_2025-11-04_at_11.11.22_AM

Additional comments from slack thread

  • Francis: I've been following this thread if it helps.
  • Chad: I suppose I should have asked Duo to give me some doc links. But with all this stuff so new and changing, I don't know how useful that might be.
  • Jessie: "In case you want to use this Agent before the UX is improved, the documentation is here: https://docs.gitlab.com/user/gitlab_duo_chat/agentic_chat/#select-an-agent"
  • Chad: So I see no mention of the @ syntax on that page. Is that actually a thing, or was Duo hallucinating?
  • Jessie: hallucination! Silly LLMs. There will be a "flow trigger" where you can @ mention a service account in an issue/MR comment to invoke a Flow from the AI Catalog. But that is for Flows (under development now), which are distinct from Agents.

Related comment on UX

I also thought it was strange that in order to use an agent from the dropdown, I had to pick the New Chat button and dropdown, even though I'm already in a "new chat". Am I starting some "nested chat"? Or just replacing my current chat? Or just throwing away my current chat?

Seems like it would be better to call that button "Use agent" or something. Or else put a separate "start chat with agent", or else make a dialog option to pick an agent when starting a chat. I don't know, not a UX designer, but this seemed strange from a user perspective.

Edited by Chad Woolley