What the official source reports
August documentation describes pre-optimization estimates and post-run token usage. Estimates neither replace invoices nor enforce spending caps.
What this means for your operation
Our view: evaluating and improving an agent also consumes resources. Pilot budgets should cover tests as well as final operations. Comparing alternatives requires a common quality baseline; cutting test costs by removing difficult cases can produce a misleading conclusion.
Next step for evaluation
Define a representative sample, measurable goal and consumption owner. Compare predicted and recorded costs after each test. Include external services in total cost. This is a Microsoft Foundry documentation update, formerly Azure AI Foundry, not a saving guarantee or general-availability announcement for every referenced capability.
Sources and references
- Azure AI Foundry — Documentation Checked on
Retrospective edition 2026-W25. The week identifies this collection’s editorial organization; the text was prepared on the date shown above.
Bring this idea into your operations.
Map the process, connect systems and evaluate outcomes with Zild.
Talk to an expert