That headline is not a hot take. It is what just started happening, in plain language, last week.
ServiceNow announced a layer called Action Fabric at Knowledge 2026. The mechanic is simple. When an outside AI agent — Claude, an OpenAI orchestrator, anything your engineering team built — wants to read data or take action inside ServiceNow, it has to pass through Action Fabric. ServiceNow meters every operation and charges the customer for each one. The unit is the action. The price scales with how hard your agents work.
This is not a one-vendor story. Workday's CEO publicly said agent metering offers "a lot of upside" for his company. HubSpot is reportedly moving the same direction. SAP took a stricter posture and is blocking unauthorized agents outright. Datadog capped agent traffic on its MCP server. Different mechanisms, one underlying recognition: the SaaS pricing model that has run on human seats for twenty years cannot pay for what AI agents do.
The press settled on a word inside five days: tollgate. The Information used it as a headline. Tech Startups, Josh Bersin, and JPMorgan's Mark Murphy all reached for the same metaphor. The word will stick because the pattern is real, and the pattern is not going away.
But the word is not the story. The word announces the story.
The story is older than it looks, and FinOps practitioners already know it.
In 2010, enterprise compute cost was predictable. You bought servers, depreciated them on a five-year schedule, and your forecast was a known number. Then AWS started selling virtual machines by the hour, engineers self-served, and the first cost surprises hit in 2013. By 2016 the surprise was an annual event in every cloud-native enterprise. A category called cloud cost management did not exist at the start of the decade. By the end of it, it was a multi-billion-dollar tooling market, with a foundation, a job title, and a routine line item on board decks.
That arc was driven by five specific powers. Every one of them is now reappearing one layer up, against SaaS.
One — the unit changes. Cloud cost shifted the spending unit from a one-time purchase (the server) to a continuous meter (the EC2-hour). SaaS is shifting from a fixed purchase (the seat) to a continuous meter (the agent action). The seat will not disappear, the same way servers did not disappear — but the variance, the surprise, and the growth all move to the meter. That is where finance teams will spend the next decade chasing the bill.
Two — the buyer moves. Cloud cost moved the spending decision from procurement to engineering, because EC2 was a self-service product. SaaS is moving the same way: the agents that drive the new bill are designed by engineers, not signed for by procurement. The bill consequence shows up after the deployment, not before the contract. Finance teams will hear about agent spend the same way they heard about cloud spend in 2014 — in retrospect.
Three — visibility lags the billing. AWS started charging meaningful money in 2011. Cost Explorer launched in 2015. Four years of darkness. SaaS is at the start of its own gap right now: no FinOps tool sees agent activity across SaaS vendors today, no SaaS vendor offers a real-time per-agent consumption API, and the customer is being asked to commit to spend they cannot measure. The first vendors that close this gap will look like wizards. The rest will keep writing post-mortems.
Four — contracts break. Annual procurement based on seat counts cannot price an unknown volume of agent operations. The remedy is the one cloud invented and SaaS has not adopted yet: pre-paid usage commits, volume tiers, burst credits, audit rights, per-agent attribution, the right to set a ceiling. New SaaS contracts will look more like cloud committed-spend agreements within two years, or they will be a finance nightmare for both sides.
Five — multi-vendor chaos is immediate. A typical enterprise runs 30–50 strategic SaaS apps. As tollgates spread, each vendor will have its own per-action unit, its own price tier, its own definition of "operation." No customer will run one vendor. No provider gains by making the cross-vendor view easy. The cross-layer picture is — as always — the customer's job to assemble.
Five powers. Different decade. Same shape.
Two things are different from cloud, and we should be honest about both.
SaaS data is more captive than cloud compute was. A workload running on EC2 can be moved to GCP. It is annoying but possible. The data inside ServiceNow, Workday, Salesforce, or SAP cannot be moved — those platforms are the systems of record for the business. Switching cost is closer to zero in cloud and closer to infinite in SaaS. That asymmetry gives SaaS vendors pricing power cloud providers never had. I expect tollgate prices to climb faster than EC2 prices ever did, and customer negotiating leverage to be thinner.
The actor is not human anymore. Cloud cost was driven by human engineers making deployment decisions. Tollgate cost will be driven by agents making real-time action decisions, often inside loops the human operator did not explicitly design. A misbehaving agent retrying a stuck workflow can spike tollgate spend in an afternoon, the way no human ever could. Cloud had idle resources; SaaS will have rogue agents. The remediation discipline is new.
These differences mean the cloud playbook is transferable, but not portable. The instincts carry over. The implementation does not.
What I expect to see in the next 24 months, said plainly.
Tollgate mechanics will be live at most data-rich SaaS vendors. The FinOps Foundation will add a formal category for agent action cost. Procurement leaders who learned committed-spend vocabulary from cloud contracts will start using it on SaaS contracts. Boards will see "cost per AI workflow" as a routine line item. The cloud-cost-era tools that try to bolt a tab onto an old data model will be in trouble; the ones aggressively rewriting their architecture will pull ahead.
Within five years, this will be as well-known a discipline as cloud cost management is today. Within ten, it will be as well-tooled as Reserved Instances are now. The arc will be faster than the cloud arc was, because the underlying market is moving faster.
If I were a FinOps leader reading this on Monday morning, here is what I would do.
I would inventory the agents already running in my company, list which SaaS platforms each one reads from and writes to, and treat that list the way I treated my first AWS cost-by-service report in 2014: the precondition for everything else.
I would call my account managers at every strategic SaaS vendor and ask one direct question — what is your roadmap for agent metering? The vendors with a clear answer will tell me. The ones that hedge are the ones I should worry about most, because the meter is coming and they have not yet decided how to charge for it.
I would start writing the next SaaS contract the way I write a cloud contract: with usage ceilings, per-agent attribution requirements, audit rights, and the right to set a budget cap before the vendor's meter does. The first vendors who balk are telling me they are not ready to support the next era of enterprise AI. I want to know that before I sign.
And I would assume that the visibility tool I am using today is not going to see this layer tomorrow.
That last assumption is the work I have spent four years building toward at Finout. A bill that keeps adding layers needs a tool that keeps adding them with it. As tollgates emerge, we are building to surface them in the same single view that already covers cloud AI services, direct AI provider contracts, AI-native tools, and embedded AI inside SaaS. The bet is straightforward — a multi-layer bill needs a multi-layer tool — and the bet is the company.
I have watched FinOps reset twice. From single-cloud to multi-cloud. From cloud to cloud plus SaaS. Both were expansions of the same underlying shape — fixed-purchase spend with usage variance on top.
This third one is not an expansion. It is the moment SaaS cost stops being fixed-purchase spend and starts being consumption spend. The seat will not vanish. It will sit alongside the meter. But the meter is now the driver of variance, and variance is what FinOps disciplines exist to manage.
The first SaaS bill that behaves like an EC2 bill landed last week.
There will be more. Soon.