AI Foundation Discovery Workshop
The demo takes ten minutes
Setting up an agent that answers questions about a PDF takes about as long as a coffee in Foundry. That same agent in production needs its own identity in Entra, knowledge sources that respect permissions, tools with well-defined boundaries, and someone on the hook for the bill. The workshop takes your case and walks the whole route.
You bring a use case
A real scenario with stakeholders and data, not a strategy paper. An agent that checks supplier invoices against purchase orders is a better candidate than a general-purpose assistant, because you can actually say what a correct answer looks like.
We sketch it out in Microsoft Foundry
We build the agent in our reference environment, not on slides: model, knowledge sources, tools, Guardrails. What Foundry can't do becomes obvious right away. What it can only do with significant effort becomes clear shortly after.
You leave with a realistic picture
What the agent can do, what it can't, which decisions need to be made before you go live, and which of the three AI Foundation entry points fits your environment. The sketch is yours to keep, even if you implement it with someone else.
How the half-day works
- Understanding FoundryUnderstanding FoundryUsing our reference environment, we show how the model catalog, Agent Service, knowledge sources, and Guardrails fit together, and where the platform stops: Microsoft delivers the control plane; it does not deliver operations.
- Breaking down the use caseBreaking down the use caseWho triggers the agent, which data does it read, what is it allowed to do, and what is off-limits. The last question is the shortest and generates the longest discussion.
- Sketching the agentSketching the agentModel selection, knowledge via RAG or MCP, tools, checks before and after every response. By the end, you have an agent that responds and a list of what still stands between it and a production system.
- Defining the entry pointDefining the entry pointWhich questions remain open, which AI Foundation entry point fits your landing zone, and what comes next if you decide to move forward. If you don't, the sketch stays with you.
The questions repeat themselves; the answers don't
Every workshop surfaces questions in the same four areas. The answers depend on your case: an agent that reads SharePoint has a different permissions problem than one that writes to SAP.
Models and agentsModel variety, ProvisionerWhich model is sufficient for your case, and how each team gets its own dedicated area in Foundry, complete with budget and boundaries, instead of a shared account for everyone.
Data and knowledgeRAG, MCPWhere the agent draws its knowledge from, and whether it inherits or bypasses the source's permissions. An index without access control is the fastest way to put the payroll in the chat.
Security and complianceGuardrails, Defender, AI GatewayWhich checks run before and after every response, how model access goes through a gateway instead of keys buried in config files, and which data must not leave the EU.
Operations and scalingBlueprint, operating model, TiersWho operates the agent once it's running, how you scale from one agent to twenty without ending up with twenty operating models, and which of the three tiers fits your landing zone.
