Date & time
17:00
Register for the panel discussion
Sign up to watch this panel discussion, hosted in partnership with Harness.
Nobody agrees on how to measure AI ROI yet, and token leaderboards have made it worse. Once token count becomes the target, engineers chase usage for its own sake. That’s part of why 63% of engineering leaders still rely on asking their teams how it’s going (LeadDev 2026 data).
You don’t need a perfect metric to get spend under control. Attribute spend to each team, compare it against an outcome measure you already trust, and set a circuit breaker that triggers a review when something looks wrong. As the policy rests on outcomes, it keeps working when the AI metric changes.
In this panel, engineering leaders share the measures they trust, the policies that have held up in their own orgs, and how they explain those policies to teams without them feeling like surveillance.
Key takeaways:
- Track AI spend by team, application, and use case, so you avoid individual leaderboards and a company-wide total that hides where the money goes.
- Compare each figure with an outcome measure you already trust, such as change failure rate or rework rate.
- Set circuit breakers that stop runaway agents and flag abnormal daily spend, with every breach triggering a review.