Agent security startup Geordie AI Ltd. today launched Cost Intelligence, a capability that links enterprise AI spending to the agents and workflows that generate it.

The launch stretches the company’s agent risk management pitch past security and into financial governance. AI spending is normally measured by consumption. Bills show which models ran, how many tokens went through them and what that came to. None of that identifies the agent behind the charge, and token use swings from one run of the same agent to the next.

Geordie put numbers to the problem. One retailer was losing $1 million a month sending tasks to models more expensive than the jobs needed. At a bank, an agent stuck in a tool loop ate 50% of the monthly AI budget in a single day.

The platform already sits close to the agent to watch how it behaves for security and governance purposes. That vantage point also exposes the economics, including what the agent did, how many tokens it used and which models it called. An operations team watching a cost line move can now trace the figure back to the agent that acted, the task it ran and the person who owns it.

Measurement happens agent by agent across cloud, code and endpoint. Geordie argues that this gives a fuller picture than tools that only see the activity routed through a single gateway or model router.

Five capabilities make up the release. Spending from every connected platform is pulled into a single view and attributed to the work behind it. Teams can drill from an organization-wide total down to platform, team, user, model, individual agent or workflow.

Caching efficiency is tracked as its own per-agent metric, flagging which agents are under-caching and what that costs. Token-efficiency mapping sets spend against volume, so a busy agent doing critical work is not mistaken for a wasteful one. Anomaly detection looks for agents in infinite loops, tasks handed to oversized models and sessions burning tokens with little to show for it.

“Tokens and invoices tell you what you consumed; they do not tell you what work drove that consumption,” said Henry Comfort, co-founder and chief executive of Geordie. “When you can see what an agent costs alongside the work it performed, you have the foundation for understanding agentic ROI, and that is exactly what Cost Intelligence delivers.”

Jacob Sweat, vice president of technology at customer Wood Partners LLC, said visibility had been one of his company’s biggest challenges with AI spending. Once the company could see what its agents cost and whether the money was buying anything useful, he said, the internal question changed from how much is being spent to whether the spend is worth it. Co-founder and chief AI officer Hanah-Marie Darley said the risk from autonomous agents “isn’t limited to security alone.”

Geordie is a venture capital-backed startup that has raised $36.5 million since its September 2025 launch. Balderton Capital led its $30 million Series A round in May. Ten Eleven Ventures and General Catalyst Group Management co-led the $6.5 million seed round before that.

Image: Geordie/GPT Image 2