Hi Glean team,
I’m seeing a mismatch between how Canvas shows up for Glean Assistant and how it behaves for custom Agents built with Agent Builder.
Summary
- In Glean Assistant, I can open a Canvas side‑by‑side and work on long‑form content there.
- In Agent Builder, Canvas is also visible as part of the authoring experience.
- But when I run a custom Agent that produces long Markdown outputs (multi‑section documents), the resulting chat message:
- Does not expose any Canvas‑related action (no “Open in Canvas” or equivalent in the UI).
- And the Agent ignores / rejects instructions in the prompt such as “open this in Canvas”, “display this in Canvas”, or “edit this in Canvas” (it just returns Markdown in the chat window).
From a user experience point of view, this creates friction: the only way to get Agent output into Canvas is to copy/paste manually, which is not obvious or scalable for non‑technical users.
Expected vs. Actual
Expected:
- For long‑form Agent responses (multi‑section Markdown documents, frameworks, or plans), I expected:
- A Canvas action in the UI (similar to export to Word/Email) to open that response as a Canvas document.
- Or, at minimum, that Agents could honor instructions like “move this draft to Canvas” by targeting Canvas as the editing surface.
Actual:
- Agent runs successfully and returns valid Markdown.
- UI only shows actions like Export to Word / Email, with no Canvas option tied to that response.
- Canvas‑related instructions inside the Agent prompt are effectively ignored: the Agent cannot open, edit, or display content in Canvas, and it only ever responds in the chat.
Repro Steps
- Open Agent Builder and create a custom Agent that:
- Reads some internal docs (via Glean tools) and
- Returns a long‑form Markdown document (headings, lists, tables).
- Save and publish the Agent.
- Open the Agent in Glean Assistant / Agents UI.
- Run a query that triggers the long‑form document output (e.g., “Generate a growth & development plan…” or similar).
- Observe the response:
- The Agent returns multi‑section Markdown as expected.
- The message actions only show export options like Word / Email; there is no Canvas / “Open in Canvas” action.
- In the same chat, ask the Agent things like:
- “Open this draft in Canvas.”
- “Edit this document in Canvas.”
- “Display this output as a Canvas document.”
- Observe that the Agent:
- Continues to respond only with chat/Markdown,
- Does not trigger Canvas, and
- Often explains that it can’t control Canvas / UI.
Questions / Request
- Is this current‑expected behavior (i.e., Agents cannot programmatically target Canvas and their outputs are not eligible for “Open in Canvas” actions)?
- If yes, is there any roadmap item or workaround to:
- Add a Canvas action for Agent responses (similar to Word/Email), or
- Expose Canvas as a destination that agents can write to (for long‑form documents and frameworks)?
This would significantly improve the experience for users who want to standardize long‑form outputs (plans, frameworks, specs) via Agents but still get the richer Canvas editing experience without manual copy/paste.
Thanks in advance for any guidance or roadmap details.