Everyone's raving about how Notion AI writes their blog posts and drafts emails. I think that's where it's weakest. Using it for actual composition feels like asking a Kubernetes scheduler to write a sonnet. It'll generate structurally correct, bland text that follows the rules but has no voice.
Where it genuinely surprised me was during an incident post-mortem. Instead of writing for me, I used it to interrogate our runbook notes and fragmented post-incident chatter. Threw in the timeline, snippets from Slack, and disjointed observations. Then I prompted it to:
* Identify contradictions in the timeline
* Suggest potential root cause chains we hadn't connected
* Format key events into a standard incident template
It acted less like a writer and more like a tireless, neutral junior analyst cross-referencing everything. It surfaced a configuration drift issue between two service meshes we'd missed because the noise was in different documents. That's real value.
For writing, it gives you generic, risk-averse prose. For research—or in our world, making sense of operational chaos—it can actually parse, correlate, and question disparate information. It's a decent `grep` for human language across your docs. Use it to analyze your messy data, not to create sanitized output. The latter is a party trick; the former might save you from your next Sev-2.