Share with your CFO
AI agents are breaking the financial logic that enterprise software has run on for a decade, and most finance functions haven’t caught up. Roi Ravhon, CEO of Finout and a FinOps Foundation board member, lays out why agentic AI spending resists every existing control: one agent run touches the model bill, the cloud bill, and multiple internal systems simultaneously, with no default way to connect those costs to a team, a product, or an outcome. Adoption metrics and token prices don’t answer the question finance actually needs answered.
What this means for your business
The split-bill problem is real and it lands squarely on whoever owns AI budgets. A single agent task can generate charges across a model provider and a cloud provider in the same second, with the engineering team seeing one job, not two invoices. Companies that are already running agents in production, rather than still piloting them, are the ones most exposed right now, because the gap between “we’re using AI” and “we know what it costs per outcome” is where margin quietly disappears.
Ravhon draws the cloud analogy correctly, though his firm sells FinOps tooling into exactly this gap, which tilts his framing toward the urgency of instrumentation over the difficulty of getting AI vendors to cooperate with it. The analogy still holds: cloud governance didn’t happen because AWS invoices became readable on their own. It happened because companies built internal tagging discipline, ownership maps, and unit cost definitions that forced accountability onto the infrastructure layer. AI adds a second cost surface on top of that, one that moves faster and sits closer to the product. A cheap model that retries five times with an oversized context window can cost more than an expensive model that finishes in one pass. Optimizing on token price alone is the wrong lever.
The gross margin math is the leading indicator to watch. Ravhon names the pattern accurately: usage looks fine for two quarters, then someone runs the numbers and the room goes quiet, because the heaviest users of an AI feature consume inference and agent time at a rate that erodes the economics of the whole product. The budget decision this reframes isn’t whether to invest more in AI. It’s whether your current cost attribution, the system that connects spending to team ownership and business outcome, is granular enough to survive the next board question about AI ROI. If the honest answer is no, that’s the renewal worth defending right now.
Concept deep-dive: Unit economics for agentic AI
Unit economics means measuring cost and value at the level of a single meaningful action, like cost per resolved support ticket or cost per merged code change, rather than at the level of total spend. For AI agents, this matters because aggregate token or infrastructure costs tell you almost nothing about whether a workflow is working. An agent that resolves twice as many tickets for three times the cost may still be the right investment. One that spends heavily and produces no measurable output improvement is waste that looks like progress until someone does the math.
Based on reporting from AI agents are changing the software business model. Enterprise finance must catch up, originally published 2026-07-16 07:18:00.
