There's a simple question I keep coming back to with Microsoft 365 Copilot: when should you Chat, and when should you Cowork?
Chat is your nimble sidekick. Rapid-fire queries against your business systems — pipeline updates, quick segmentations, lead conversions, "what changed since yesterday?" It's fast, conversational, and perfect for the transactional moments in a working day.
Cowork is a different gear. When it's time to go deeper — automating follow-ups, drafting a pitch deck, building an RFP response, producing a live dashboard — you shift into Cowork mode and let the agent do the heavy task work. Not a summary. A finished deliverable.
What excites me is that both can run on the same MCP layer. One set of plumbing powers the everyday transactional work and the heavy lifting — and it ports to whatever back-end you run.
Five plugins that turn live data into deliverables
To push Cowork past "working with only my data" into AI that actually does the work with my apps, I built five plugins on top of that MCP layer — for the systems workflows and data already live in: ServiceNow, Salesforce, Jira, Workday, and Coupa.
My favourite is the ServiceNow one. It writes my weekly IT incident report as a live HTML dashboard, schedules itself to run every morning on its own, and helps me improve the service over time. The demo I love: a declarative agent orders a laptop straight through the MCP server using MCP Apps, then Cowork generates the weekly IT review as a dashboard, then schedules the daily pulse to run automatically.
Same plumbing, different jobs — real dashboards, decks, Teams cards, Excel workbooks and reports you could send as-is. And it's all open source, so you can take it apart and point it at your own data.
This is the pattern I think matters: don't force everything through chat, and don't rebuild integrations for every use case. Build one governed MCP layer, and let Chat and Cowork each do what they're best at.