Full disclosure, same as my earlier post: I build Mermaid Plus at NGPILOT, so I have a horse in this race — which is exactly why I'm asking instead of assuming.
Here's the pattern we keep hearing about:
A team writes an architecture page in Confluence. Text gets updated constantly — AI teammates summarize, rewrite, draft. But the diagram on that page is a PNG exported from some tool, so it slowly lies: two services renamed, one queue added, nobody re-exports. New joiners onboard against a diagram that's three sprints stale.
The text on the page is fresh. The most information-dense thing on the page is the most out of date.
What I want to learn from you:
- Does your team have a process that actually keeps diagrams current — or is "the diagram is outdated" just accepted background noise?
- If you solved this: was it code-as-diagram (Mermaid/PlantUML in source control), a dedicated tool, review rituals, or something else?
- Has anyone tried letting an AI agent do the diagram updates, and where did it break?
I wrote up our own take on why text-based diagrams are the format AI-era Confluence needs — with what an agent operating diagrams (not just generating images) looks like in practice — over in this post. But that's one vendor's view, and I'd genuinely like to know how teams are handling this today, including "we don't, and here's why it hasn't mattered."
Every answer helps us build a better agent — even the ones that say the whole problem is overblown.
