Have you already played around with the Agent Tools asset in Studio? It lets you expose your governed Celonis data - KPIs, records, insights - through tools that an external AI agent can call via the open MCP standard. So instead of guessing, your agent just asks Celonis.
My favourite little detail: the description you write for each tool is what the agent actually reads to decide whether to use it. So treat it like a mini prompt, not a label. Write it well, and the agent picks exactly the right tool, first time. β¨
(And by the way, if you want the full end-to-end: we just launched a course on exactly this π)
What would you let an agent reason over first? Comment below!
