The number and what sits behind it
On July 1, Gartner published a figure designed to be repeated in board meetings: up to 234 billion dollars of enterprise application software spend is at risk from agentic AI between now and 2030, which the firm pegs at roughly 20 percent of enterprise SaaS spending by the end of the decade. The mechanism has a name, agentic arbitrage, and it describes what happens when AI agents complete tasks across several systems and collapse the need for humans to log into each one. When the human stops opening the application, the seat that the application was priced on stops being worth paying for.
George Brocklehurst, a Managing Vice President at Gartner, framed the shift bluntly, saying agentic AI changes the economics of software because agentic systems deliver outcomes directly and make the underlying application invisible. His sharper point for buyers is a change in what they are purchasing at all. In his words, you are no longer buying software primarily for people, you are increasingly buying it for agents. That reframing matters because a company that provisions software for a fluctuating population of agents cannot forecast cost the way it did when a seat mapped to an employee.
Why the seat model is unwinding
For two decades the per-seat license was the cleanest metric in enterprise software, easy to forecast and easy to grow as headcount grew. Agentic automation breaks that arithmetic by decoupling value from the number of humans logged in. Gartner expects at least 40 percent of enterprise SaaS spending to shift toward usage, agent, or outcome-based pricing by 2030, with seat-based revenue falling from around 21 percent of the market to roughly 15 percent. Vendors see the same trend line their customers do, and they are repricing ahead of it rather than waiting to be disintermediated by the agents running on their own platforms.
The transition transfers forecasting risk from the vendor to the buyer, which is the part CIOs feel first. Sidharth Ramsinghaney, Director of Strategy and Operations at Twilio, put the operational reality plainly, saying the most immediate problem is budget volatility that most organizations have never had to manage before. A fixed annual seat cost becomes a consumption meter that scales with usage and, in many contracts, without a ceiling. Finance teams that could plan software spend to the dollar now face a line item that moves with how hard the agents work in any given month.
Installed-base harvesting is the quiet driver
AI feature bundles are the public justification for the current wave of price increases, and the underlying driver is installed-base harvesting: extracting more revenue from customers who are expensive to move. New-customer growth has slowed across enterprise SaaS, so most vendors' 2026 revenue plans depend on lifting the value of accounts they already hold. Industry data shows average enterprise SaaS spend has climbed to about 55.7 million dollars annually, up 8 percent year over year, while application portfolios have stayed roughly flat at around 305 applications. The growth is coming from price, AI tiers, and consumption charges rather than from new tools entering the estate.
Microsoft offers the clearest recent example. From July 1, 2026 the company removed volume-based discount tiers from Enterprise Agreements for online services and raised prices on select Microsoft 365 suites, with reporting from licensing specialists putting the effective increase for large customers between 6 and 12 percent across E3 and E5 plans. AI pricing uplifts elsewhere in the sector run from 20 to 37 percent. Whatever the mechanism, bundling, hybrid pricing, evergreen renewal clauses, or shorter terms, the objective is the same, which is growing average contract value inside accounts the vendor already controls.
The new lock-in is knowledge, not just data
The lock-in enterprises understand is technical: data gravity, workflow dependency, and the web of API integrations that assume a vendor's data model. Gartner names a second form that is easier to miss. As agents run inside a vendor's platform, the operational learning they generate, which workflows work, where exceptions arise, how a process actually behaves, can accrue to the software provider rather than to the customer. Gartner calls this a Knowledge Retention Rate concern, and it represents a competitive risk because the enterprise can end up renting back insight into its own operations.
This changes what contract diligence has to cover. Brocklehurst's advice is to scrutinize the contract as much as you scrutinize the technology, because clauses restricting autonomous or agentic use, and clauses governing who owns the learning an agent produces, now carry strategic weight. Jasper Geurts, CTO of Software Improvement Group, adds that the token economy is opaque to the CIOs he talks to, which compounds the problem. A buyer who cannot see how consumption is metered and cannot claim the knowledge their agents generate is negotiating from a weak position on both cost and control.
What the renewal conversation becomes
The practical consequence is that software renewal stops being a procurement exercise and becomes an architecture decision. When pricing moves to consumption, the cost of a workflow depends on how it is designed, so the CIO who understands where agents will run and how often gains real leverage at the table. Maksim Hodar, CIO of Innowise, advises piloting new pricing models before scaling them, which is sound because a consumption contract signed without a usage baseline is a blank check written against a meter nobody has read yet. Modeling the agentic workload before committing is now part of the negotiation, not a follow-on task.
We would treat the next renewal cycle as the moment to reprice the relationship deliberately rather than reactively. That means demanding usage caps or predictable tiers, securing rights to the operational data and learning your agents produce, and pressure-testing every AI bundle for whether it delivers value proportional to its uplift. The 234 billion dollar figure is a projection about vendors, and the reason it belongs on a CIO's desk is that the same shift decides whether your software budget stays a manageable fixed cost or turns into a variable expense you no longer fully control.
The build-versus-buy question resurfaces
When outcomes can be assembled by agents orchestrating across systems, the application layer that used to be the obvious purchase starts to look optional for some workflows. That does not mean ripping out core systems of record, which remain the safest place to keep data and process. It means the mid-tier tools priced on seats that agents can now bypass are the ones most exposed, and the ones where a CIO should ask whether an internally orchestrated agent could deliver the same outcome at a fraction of the license cost. The arbitrage Gartner describes cuts in the buyer's favor when the buyer is the one running the agents.
The vendors moving fastest to usage and outcome pricing are, in effect, conceding that the seat was always a proxy for value that agents now measure directly. For the reader planning a 2026 and 2027 roadmap, the discipline is to separate the systems worth their rising price from the ones being harvested, and to reallocate the savings toward the orchestration and governance layer that makes agents safe to run in production. The repricing is happening whether or not any single enterprise is ready, and readiness here is mostly a matter of knowing your own workflows better than your vendors know them.
