Beyond Tokens: Key Takeaways from Tokenomicon & FinOps X
About the Author: Lucas Paratore - Product Manager, CloudHealth
Last week in Amsterdam marked the first officially combined Tokenomicon and FinOps X event, with around 500 people in person and more than 2,000 joining virtually. You might have expected a conference called Tokenomicon to be about tokens and optimizing token usage, but that isn't what happened. The conversations were about visibility into AI costs, how involved the FinOps team should be, and how an organization decides whether an agent is creating enough value to keep paying for it.
Tokenomics is bigger than Tokens
During one of the lighting talks, the speaker mentioned they prefer "AI Economics" to "Tokenomics", and I think that captures the sentiment of the whole event. Depending on where you sit, Tokenomics means something completely different, and rarely has to do with the tokens themselves. Engineering focuses on gateways and model selection, finance thinks about forecasting and unit cost, and business leaders have to determine value and understand the total cost.
FinOps role in AI strategy
Throughout the sessions, it became clear that the FinOps team must be actively involved in the conversation as organizations develop their AI strategy. During Dean Oliver's chalk talk, "Agentic AI in FinOps: From Insight to Autonomous Action", the room generally agreed that troubleshooting an agent stuck in a loop is outside the FinOps team's ownership. They should, however, be responsible for surfacing it when it happens and getting it in front of the right engineer.
Beyond engineering, the FinOps team also needs to close the gap with procurement. Billing can be complicated to track as teams buy both directly from model providers and through cloud marketplaces. That makes it critical for an organization's purchasing strategy to align with its technology strategy. The FinOps team is well positioned to own that alignment, because it already sits between what engineering consumes and what the business has committed to.
Budgets tied to value
One of my favorite slides came from Booking.com. Rather than granting bigger AI budgets because usage went up, teams had to earn them. That meant consistent usage over four to six weeks instead of a single spike, quantifiable proof like hours saved or increased output, confirmation that the usage wasn't just inefficient prompting or the wrong model, and a tie back to a Jira Epic. Creating this culture both pushed teams to get more value from their AI spend and made that spend easier to forecast.
Agent-to-agent chargeback
As organizations mature, their agents start working across business units rather than inside them, and that makes chargeback much more complicated. In an agent-to-agent system, a FinOps agent might call a Jira agent owned by engineering. The work was initiated by FinOps, but the cost lands on engineering's bill. This sort of shared cost problem has already existed with cloud for years, except AI costs can climb much faster. FinOps teams must decide how agent-to-agent costs are allocated before these systems spread.
Still a FinOps Conference
Although it was called Tokenomicon and some of the keynotes were aimed at a more technical audience, most of the conversations at the chalk talks and around the floor were FinOps conversations. People wanted to know how to get visibility into their AI spend, who should own it, and how to justify continuing to pay for it. Those are the same problems FinOps has been solving for cloud over the last decade. While AI adds new variables and the costs move faster, it is ultimately still a FinOps problem to solve.
As AI is making its impact on FinOps, how is your team responding? Are you prepared to manage your organization's AI footprint? Partner with a team that will walk with you every step of the way.