Cut AI Agent Costs with TokenOps

Why watch

Gareth Bland, Chief Data Scientist at Microsoft, explains TokenOps: the practice of optimizing where token spend delivers maximum value in AI agents. The video covers why reasoning models make output tokens expensive, how agent costs grow quadratically with more turns, and how caching can reduce costs to a tenth. He introduces three measurement ratios (autonomous completion rate, token efficiency ratio, value generation ratio) and an S-curve to determine if your team is underinvesting, getting real leverage, or over-engineering. For engineering leaders and developers building AI agents who need to manage costs effectively.