There's a whiteboard in almost every Confluence space I look at. The architecture one, usually. Projected in design reviews, opened by every new joiner, corrected live during arguments with a borrowed mouse. It's the most human artifact in the space — and the most expensive to keep true.
Confluence now has genuinely good AI for pages, Jira, and chat. So I tried the obvious thing: I asked AI to update that whiteboard. Not to create a new one — to maintain the one that exists. Rename a service, regroup a few boxes, find the other boards that have gone quiet.
Here's what I found, with receipts, and what we shipped to change it. I'm Simon from NGPILOT — we build Excalidraw Plus Whiteboards for Confluence, so read everything below as an interested party. But the starting facts aren't ours; they're Atlassian's own documentation.
What official AI does on whiteboards today
Credit where it's due: Atlassian shipped real AI for whiteboards. The official documentation is specific about what it covers.
Creation. From the Getting started with whiteboards guide: "Turn a text prompt into new whiteboard content using Create with Rovo or Smart Create for whiteboards." Describe what you want, get a fresh board. That's a real capability, not vaporware.
Summarization. The Summarize a whiteboard using Rovo doc: "Use Atlassian Intelligence to generate a summary of whiteboard contents." A board can be read back out as text — stickies become a page.
So the official AI feature list for whiteboards has two verbs: create and summarize. AI can put new content onto a board, and it can read a board back out.
What's not on the list is the third verb — the one that whiteboards, more than any other Confluence artifact, desperately need: maintain. No "update the board to match this change." No "rename Service A everywhere it appears." No "which boards in this space are six months stale?" Once a board exists, the AI steps back out, and the board reverts to being maintained the way it was in 2019: by hand, by whoever notices, eventually.
Meanwhile, the rest of Confluence is having its efficiency revolution
This isn't a complaint about the whiteboard feature list in isolation — it's that the gap sits against this backdrop. In 2026, the AI-assisted workflow is the workflow. Pages get drafted, summarized, and translated by Rovo. Jira issues get created from meetings and chats. Automation plus agents handle the follow-ups nobody wanted to do by hand. The text side of Confluence has been thoroughly rebuilt around "describe the outcome, review the result" — and that loop is genuinely transformative: work that was forty minutes of typing becomes one prompt and a review.
The whiteboard got the first half of that revolution (prompt in, board out) and the last half (board in, summary out). It missed the middle — and the middle is where diagrams actually live their lives.
Why the gap is structural, not just a missing feature
It's worth being precise about why, because it explains why nobody else has filled it either.
A whiteboard looks like a drawing. Native whiteboard content isn't exposed through the public API as elements you can read or write — the Confluence REST API v2 covers whiteboard objects and their properties (title, space, metadata), not the shapes, connectors, and text on the board. There is no element-level read/write surface for an agent — or anyone else — to operate on. AI can generate a whole new board (generation happens inside Atlassian's own systems) and summarize one (reading happens there too), but there's no door through which any agent, Atlassian's or third-party, can edit what's already on the board.
And the board needs editing more than it needs anything else. The first version is the cheap part. The architecture drifts, the team renames services, the flow changes — and the board starts lying, one box at a time. The failure mode is the oldest one in Confluence: a picture of a decision that used to be true. Everyone knows AI is supposed to be fixing exactly this class of busywork. On whiteboards, it can't even see the work.
So how should a whiteboard work with AI agents?
That's the question I keep coming back to. The answer can't be "abandon whiteboards for text" — whiteboards are where teams think together, and the artifact the room argues on is visual by nature. The answer also can't be "wait for the roadmap."
Our answer is the one we could build: a whiteboard whose structure is exposed, with an agent standing on it. An Excalidraw board is stored as a scene — a structured list of elements (rectangles, arrows, text) with types, coordinates, and properties. Humans drive it through a canvas; underneath, it's data a language model can read and rewrite. That's the door the native whiteboard doesn't have, and it's why we built Excalidraw Plus Whiteboards for Confluence around a Rovo agent that can actually operate the board:
- Maintain what's there — "rename Service A to Auth Service," "group the three AWS boxes and make them green," "move the queue next to the ingestion service." Element-level edits to an existing board, which is precisely the verb the official feature list stops short of.
- Create — architecture, flowchart, sequence, mindmap, kanban, or wireframe boards from a prompt or template, dropped onto the page as real editable shapes.
- Find — "which pages in this space have diagrams?" across a whole space, so the stale boards get discovered before they mislead someone.
- Convert — Mermaid or PlantUML into Excalidraw shapes, and Excalidraw back to Mermaid. Real, editable artifacts on both sides — when a board needs diffing or versioning as text, one prompt moves it over.
The guardrails matter as much as the verbs. Every agent write lands as a normal Confluence page version — attributable, revertible from page history, no "backup copy" rituals needed. The agent re-checks the latest saved version before each write so it can't clobber a teammate's mid-conversation edit, and it refuses to delete one of several similar shapes without confirming which one you meant.
The honest footnote
Two things we won't claim. First, "easy to edit" and "kept accurate" are still different problems — an agent that edits on request doesn't notice that your architecture changed; the diagram owner and the PR checklist still own that. Second, auto-layout today is top-to-bottom flow or grid, and nothing fancier. If your diagrams lean text-native, our sibling app Mermaid Plus runs the same kind of agent on that side, and the convert bridge works both ways.
But the short version of this article: Confluence whiteboards got Create with AI and Summarize with AI. The verb that matters over a board's real lifetime — Maintain with AI — is missing from native whiteboards, and structurally can't be added by anyone while board content has no agent-facing surface. Excalidraw Plus gives whiteboards that missing verb today.
So, the question back to this community: if an agent could maintain your whiteboards, what's the first maintenance task you'd hand it? And for the Atlassian folks reading: is maintain on the whiteboard AI roadmap? The comments are open — I'd genuinely like to know which way both answers go.
