You’ve built the agents. You’ve deployed the copilots. Maybe you’re running local models through OpenClaw and Ollama alongside Azure OpenAI endpoints. For two decades, Microsoft sold work software the way it sold seats — one user, one license, one predictable line on the budget. That model just changed. As of June 2026, Copilot Cowork bills like Azure, not like Office: every task meters the model it picked, the context it pulled, the tools it called, and how long it ran. GitHub Copilot made the same move. The agentic layer of your stack is no longer a seat you buy, it’s a meter that runs every time an agent acts on your behalf.

Most organizations will discover this the way you discover fuel burn on a long day on the water: at the dock, looking at the receipt, wondering where it all went. This session is about governing those costs before they escalate — where AgentOps meets FinOps for the hybrid AI era. We’ll share patterns from real environments balancing Azure-hosted models with local alternatives, show how model routing becomes a deliberate cost lever instead of a default, and walk through a cost-assessment framework that helps leaders make build-vs-buy-vs-run-locally decisions they can defend to Finance. Because AI success was never measured by how many agents you deployed, it’s measured by whether you can govern them like the material business expense they’ve quietly become.

Three takeaways

Govern before it escalates — a build-vs-buy-vs-run-locally framework and the cost visibility to catch runaway spend early
Spot the cost drivers — the per-task cost anatomy of Copilot Credits, Cowork, and Work IQ, and where spend actually hides across a hybrid portfolio
Route by cost — when Anthropic, when GPT, when a local model through OpenClaw or Ollama earns its keep

Tokenomics: Governing AI Costs Before They Escalate

Other Events