For thirty years, business software has been priced on one assumption: work is done by humans, humans log in, so you charge per human. Count the heads, multiply by the sticker price, send the invoice. The seat was never really the product. It was a proxy for the amount of work flowing through the tool.
AI agents break the proxy. An agent does not log in. It does not have a seat, a session, or a monthly active user profile. It calls APIs, reads and writes records, and executes the workflow directly, and it can do the work that previously justified several licensed humans across several tools. Gartner now estimates that up to $234 billion in enterprise application software spend, roughly 20% of the market, is structurally at risk by 2030 because of exactly this. Gartner calls it agentic arbitrage, and it is the most consequential change to how software is bought since the move to the cloud.
To be clear about what that number means: it is not a prediction that $234 billion of software disappears. It is a prediction that the commercial model behind that spend stops being defensible. The software keeps running. The seats stop making sense.
The seat is already dying
This is not a 2030 story. Seat-based revenue has already slipped from about 21% to 15% of enterprise software contracts in roughly a year, and Gartner expects at least 40% of enterprise SaaS spend to move to usage-, agent-, or outcome-based pricing by 2030. Analysts covering the research put the vendor transition window at three to four years.
That window cuts both ways. Vendors have three to four years to change how they charge you. You have the same three to four years to change what you agree to pay for.
What agentic arbitrage looks like in practice
Take a routine support workflow: a ticket arrives in the helpdesk, someone checks the customer record in the CRM, verifies the invoice in the billing tool, drafts a reply, and updates all three systems. Today that flow can justify three seats per support person, because a human has to click through three UIs.
An agent runs the same flow through APIs. The helpdesk, CRM, and billing tool all still do their jobs, but the human UI interactions that justified the per-seat licences largely vanish. When your usage report shows that ten people used to touch a tool daily and now two do, the renewal conversation changes, and you are the one holding the data.
The exposure is not evenly distributed. The categories most at risk are the ones whose main value is workflow coordination, data entry, reporting, and shuffling data between systems. The least exposed are systems of record with proprietary data and tools whose value is deep analysis rather than clicks. A useful question for every line of your software budget: is this tool the system of record, or is it the glue between systems? Agents eat glue first.
The four models replacing the seat
What replaces per-seat is not one model but four, and you will likely see all of them on invoices within a year or two.
1. Usage-based. You pay for what runs: tasks, API calls, compute, tokens. Honest, but variable, and as we showed in the cheapest model on paper can cost the most, usage cost is a distribution, not a sticker price. Budgeting for it requires per-task visibility.
2. Per-agent. The seat comes back wearing a robot costume: you pay per active agent instead of per human. Simple to budget, but watch the definition of "agent", because a vendor that counts every automation as a separate agent has just reinvented seat inflation.
3. Outcome-based. You pay per resolved ticket, qualified lead, or processed invoice. The most aligned with value and the hardest to audit; you will want your own numbers on what an outcome is worth before you sign someone else’s.
4. Hybrid. The most common near-term pattern: a base platform fee plus metered agent activity. The trap to watch is the cover-charge contract, where the familiar-looking seat fee merely grants access while the real, growing bill is the metered AI activity stacked on top. If your renewal has both a per-seat line and a credits line, price the credits line as if it were the whole product, because within two years it will be.
The buyer’s playbook for the transition
You do not need to predict which model wins. You need four habits before your next renewal cycle.
Audit seats against actual UI usage. Most teams pay for licences nobody has logged into for months, and agents will widen that gap fast. The usage report is your negotiating leverage; pull it before the vendor does.
Price work in cost per outcome, not cost per seat. "$40 per user per month" tells you nothing once agents do the work. "$0.30 per resolved ticket" is a number you can compare across a human workflow, a SaaS tool, and an agent. It is the same discipline as the success-criteria question in why 40% of AI agent projects get cancelled: pick the number before the invoice picks it for you.
Demand per-task cost visibility from anything metered. Usage pricing without instrumentation is how team AI spending gets away from you. If a vendor meters your usage but cannot show you cost per agent and per task on one screen, the meter only works in one direction.
Renegotiate at renewal, not after. The three-to-four-year transition window means every renewal between now and then is a chance to move a contract from seats you do not use to usage you can measure. Sitting still is its own decision, and as we argued in the real cost of not adopting AI, it is rarely the cheap one.
Where Crewdle fits
Crewdle was built on the side of this shift that buyers should be on, which makes the pitch mercifully short.
- No seats, anywhere. Crewdle is pay-as-you-go: credits buy work, not logins. An agent that runs little costs little, and adding a teammate does not add a licence.
- The meter is on your side of the glass. Crewdle Admin shows spend per agent and per task on one screen, so usage pricing comes with the instrumentation that makes it safe to budget.
- Agents without the plumbing. Crewdle Connect runs the agents that do the work seats used to represent, so the arbitrage happens in your favour rather than your vendor’s.
- Run your own numbers. The savings calculator turns "cost per outcome" from a slogan into a figure you can put in front of a finance lead.
The takeaway
Per-seat pricing priced human attention, and agents remove human attention from the loop, so the model was always going to break; Gartner has just put a number and a date on it: $234 billion by 2030. The winners of the transition will not be the companies that guess the right successor model. They will be the ones that walk into every renewal knowing their seats-versus-usage numbers, price work in cost per outcome, and refuse any meter they cannot see. The seat counted people. What comes next counts work, and that is a number worth knowing before your vendors know it better than you do.
Start for free, run one bounded workflow pay-as-you-go, and see what your work actually costs when nobody is charging you for the chairs.
Ask about this article 
