AI systems are powerful, but they reason slowly and cost a lot, even when the task is only a simple decision. Should this request be approved? Which agent should handle this ticket? Does this answer need human review? Using a full LLM to answer these questions adds latency and cost. A decision model is built for exactly this job: it reads the input and returns a calibrated probability for each option, so your application knows when to act, when to escalate, and when to keep a human in the loop.

In this hands-on session, we explore Microsoft-Decision-1, Microsoft’s new decision model, available in Microsoft Foundry. It follows JEV, the model that opened this category in September.

What we will cover:

-What decision models are, and how they differ from LLMs?
-Why every major AI company is now building decision models?
-The fast-moving evolution of decision models, from JEV (September 15) to Microsoft-Decision-1 today.
-What these new type of models change for AI engineer's: how we design, build, and pay for AI workflows?
-Using Microsoft-Decision-1 inside a LangGraph workflow, with the graph and agents ready-made so you can focus on how decisions route through the edges.
-Monitoring and tracing the application with LangSmith.