We've been building a multi-step RFP Proposal Builder agent and ran into a UX challenge. The agent walks users through a structured, feedback-driven workflow, extracting RFP objectives, categorizing them, designing an outline, and mapping it to standard content blocks, before producing a full proposal document. Each step requires the user to review, tweak, and confirm before moving forward.
The problem we're hitting, two related pain points:
- No canvas / split-pane editing. When the agent generates or refines proposal content, there's no side-by-side editable view. The user has to scroll back through the conversation to re-read what was generated, which breaks the review loop β especially for long, multi-section documents. Glean Chat itself supports this with a canvas view and tools like ChatGPT and Claude have similar split-pane models. but this capability doesn't carry over into the agent workflow.
- Every iteration creates a new document. Each time the user provides feedback and the agent refines the output, a brand new Google Doc is created. After a few rounds, the user is left with multiple fragmented versions and has to manually reconcile them. What we really want is a single living document that gets updated in place across the session.
We understand a full canvas model may not be supported yet, and we get that iteration over content is still possible, just not in a canvas-like model. But the "new doc every iteration" problem feels like something that could be addressed even without a canvas UI, by having the agent track and update the existing document ID across turns.
Our questions:
- Is there any canvas or split-pane editing capability in agent workflows today β even in beta?
- Is there a recommended pattern to have agents update an existing document rather than always creating a new one?
- Is either of these on the roadmap, and if so, what's the timeline?
Would love to hear if any one have run into similar friction with document-generating agents, and how you've worked around it.
