An enterprise AI assistant deployed on your cloud, with your employees' identity. Then AI at work on your own documents, not on a demo.
A pilot starts with a shared API key. Nobody knows any more who sent what, or to which model.
The tool answers general questions well. It knows nothing of your processes, your logs or your architecture diagrams.
You automate the process written in the procedure. The one teams actually follow is somewhere else.
Claude Desktop and Claude Code in third-party mode, connected to the client's Microsoft Foundry. Named Entra ID authentication, no API key. Agents, tools and MCP connectors set on that foundation.
I read all the process documentation, I watch people work, I ask them a few questions, I open the applications and the traces in read-only mode. Out comes the real workflow, and the places where AI belongs in a human workflow.
The practices that work become shared skills and methods, executable in the value stream. You move from AI in a human workflow to humans in an AI workflow.
I work under the inspearit banner. The choice of tool depends on your context.
Built on your documents, your logs and your diagrams, and a foundation your teams already work on.
In the field. At a large enterprise, a mixed team of consultants and in-house staff put this foundation into service on 3 workstations so the AI could read internal documentation, logs and architecture diagrams.
No gain, no rollout.
30 minutes is enough to place your starting point: what is already there, what the AI can read, where to begin.