AI agents will make counting SaaS seats look outdated

A seat measures access rather than completed work

Consider a hypothetical British engineering consultancy preparing a quotation. Someone checks previous jobs, finds an agreed rate and drafts a proposal. A colleague approves it, then another person updates the customer record.

An agent could be assigned the preparation work, subject to access controls and approval before anything reaches the customer. The consultancy would still need trustworthy records and a place to retain the approved quotation. But fewer people might need to spend their day inside each application.

This separates two buying decisions: how to keep business information dependable, and how much effort it takes to use that information.

A per-user price can remain sensible where people collaborate extensively. It becomes less persuasive when an agent takes over repetitive handling. Automation does not have to eliminate jobs to affect licences. Staff could spend more time advising customers while needing less direct access to particular tools.

BCG’s August 2025 analysis of software pricing identifies this pressure explicitly. It examines resource consumption, interactions and completed jobs as alternative charging units, while expecting hybrid arrangements to combine new measures with established subscriptions.

For a UK finance director, none of those alternatives automatically represents a better deal. Paying for each action can reward activity that produces little benefit. A completed-job charge can create an argument over whether the job actually finished. A flat subscription can be worth keeping because next month’s cost is known.

The useful question at renewal is which unit reflects the work being bought, with enough predictability to run the business.

British customers are meeting agents inside existing platforms

The British customer accounts in Salesforce’s December 2024 account of British Agentforce customers show how agents can enter an established service. Bionic described using them for meter readings and switch tracking. Secret Escapes described automating routine travel-service tasks, including cancellations and booking updates.

These are customer statements published by the supplier, rather than independent performance tests. Their value here is the description of the work being attempted. Both connect agents to an existing service and its customer information.

That gives British buyers a practical starting point. Examine the repetitive work already passing through a system the business trusts. Establish whether the supplier’s agent can perform it acceptably, then compare that option with an external service or a process change.

The international suppliers’ pricing announcements show how they intend to sell that capability. In September 2025, Workday introduced Flex Credits as a subscription-based consumption model, with an initial allocation inside existing subscriptions and additional credits available as usage grew. Its announcement of Illuminate agents and Flex Credits describes a recurring software relationship with a new charging mechanism attached.

HubSpot took a similar route. In May 2025, it announced that Breeze Customer Agent would become available across its Professional and Enterprise offerings through HubSpot Credits. The announcement described a usage-based model attached to an agent embedded in its existing platform.

A supplier holding the customer relationship can introduce automation where the customer’s information already sits. The buyer can assess an additional capability without immediately replacing the surrounding application.

The subscription therefore survives, at least in these models. For UK customers, the negotiation is whether the new charge buys enough useful work to justify itself, and whether existing access commitments can fall as that work moves to agents. An incumbent can lose daily human attention while retaining the account. That does not settle what the account should cost.

Britain’s opportunity lies in knowing the work

Sage’s Making Tax Digital announcement shows what a specifically British agent proposition can look like. In September 2025, the company outlined an agent intended to organise client information, set reminders and chase documents around quarterly updates. Its MTD agent announcement placed those capabilities within the Sage platform and its UK accountancy offering.

Read that as a product direction announced at the time, rather than proof that every promised function subsequently reached every customer. Even with that distinction, the commercial proposition is clear: Sage was proposing to perform more of the work surrounding its records.

For an accountancy practice, document chasing is recognisable work. A useful agent has to know which client has supplied which information, whether a reminder is appropriate and when an accountant needs to intervene. Fluency alone cannot establish those conditions.

Specialist entrants are approaching the same problem from a different position. In a July 2025 interview with Tech.eu, Gradient Labs co-founder Dimitri Masin described building agents around the standard operating procedures used in customer operations. Founded by former Monzo employees, the company focused on regulated businesses, including financial services.

That focus points to an opportunity for British technology firms. Experience of how a particular service handles exceptions can become part of a product. A developer who understands disputed payments or missing accountancy records has something specific to test an agent against.

I would put greater weight on that knowledge than on a claim that UK rules provide permanent protection from overseas competition. Competitors can hire specialists. Customers can change suppliers. A defensible business has to keep demonstrating that it handles the work well.

Services companies also have a role. Kainos announced in June 2025 that it had become a design partner for Workday’s AI Agent System of Record. That offers a concrete example of a UK firm participating in agent development through an established international platform.

For British consultancies and managed service providers, the opportunity extends beyond installing an agent. Someone has to maintain its connections, test changes and investigate the cases it cannot complete. Those responsibilities could support recurring service revenue, provided the supplier can deliver on a more demanding promise than access to a tool.

A British business still needs dependable records

The prospect of reducing licence costs should not lead UK buyers to confuse replacing an interface with replacing a business system. Klarna’s experience provides a useful correction.

In March 2025, diginomica reported Sebastian Siemiatkowski’s explanation of Klarna’s software consolidation. He described bringing information together through an internal technology stack that included Neo4j, then giving AI access to that knowledge. He explicitly rejected the interpretation that Klarna had replaced SaaS with a large language model.

Consolidation required building and organising a replacement. Removing a supplier’s invoice did not remove the need to maintain software.

For the hypothetical engineering consultancy, an internally built agent might demonstrate the normal quotation process convincingly. That would not establish what happens when two employees change a record at once, a connection stops working or a customer disputes the result. Whoever owns the replacement inherits those questions.

IDC makes the broader counterargument in its September 2026 assessment of the SaaS market. It argues that agents depend on structured business data and dependable processes, and reports enterprise application growth near 13% through the first half of 2026. That market measure is broader than SaaS subscriptions, but it offers evidence against a present collapse in demand for business applications.

I find this the strongest objection to the replacement thesis. A system of record earns its place by preserving an authoritative account of what happened. An agent still needs somewhere dependable to record an approved transaction.

That continuing need does not guarantee the survival of every supplier currently providing it. A British customer could consolidate records elsewhere, rebuild a narrow process or buy a better-supported alternative. The work remains; the contract is contestable.

Less visible applications could change the renewal negotiation

Imagine the consultancy’s staff asking one assistant to prepare a quotation, check an overdue invoice or find the latest project decision. If the assistant can obtain the right information and act through approved connections, staff may have less reason to learn each supplier’s interface.

That future is conditional. Access rights and commercial terms must permit it, and the connections must work reliably. A conversational front end cannot remove those requirements.

If those conditions are met, British buyers could increasingly judge a supplier on how well another system can use its capabilities. Clear documentation and predictable charges would become part of the product’s usefulness, alongside its screens.

Buyers could gain room to replace narrow tools without retraining everybody. They could also become dependent on the supplier controlling the assistant. Whoever maintains the accumulated instructions and exceptions may become expensive to leave.

British companies therefore need to examine portability at both levels. Can they export their records? Can they also retain the instructions, evaluation cases and permissions that make their agent useful?

It would be an expensive mistake to escape one restrictive contract by building the next dependency above it.

Paying for outcomes makes the contract harder

For a UK business considering an agent proposal, outcome pricing has an obvious attraction: pay for a problem to be resolved. Agreement becomes harder when the customer returns the next day.

Was the original interaction successful? Does the second contact count as another charge? What happens when the agent completes its assigned step but leaves a colleague with corrective work?

BCG’s pricing analysis documents disputes over whether support issues were truly resolved. A result-based charge needs an agreed definition and a way to challenge the bill. Buyers need evidence behind the charging event. A closed ticket records an action; establishing whether the customer’s problem was solved requires a separate test.

There is also a cash question. Time saved does not automatically become money available to spend elsewhere. If employees use freed capacity to improve service, that can justify the purchase. It needs to appear honestly in the business case as additional capacity or better service, rather than a payroll saving that never occurs.

ServiceNow’s Kainos customer story illustrates the distinction. It reports average case-resolution times falling from seven hours to two within a broader deployment involving Customer Service Management Pro and Now Assist. Kainos describes an ambition to serve more clients without increasing headcount at the same rate.

The case supports examining capacity per employee. It does not isolate the contribution of an autonomous agent or establish that another buyer will achieve the same result.

For a British company, the calculation must include the work retained by staff. Reviewing exceptions has a cost. So do maintaining connections and running the old service during a pilot. A lower licence subtotal cannot establish a lower total cost.

The counterargument that could preserve the seat

Per-user subscriptions retain a strong defence where collaboration is the product. The consultancy’s engineer reviewing a drawing needs access. So does the colleague approving it. An agent preparing information does not necessarily remove either person’s need to inspect the result or understand the decision.

A fixed subscription can also let a company experiment without treating every action as a spending decision. Metering can discourage useful activity if employees are uncertain about the bill. For a small British firm without dedicated cost-management staff, a predictable contract may be the better purchase even when a usage model looks cheaper in a demonstration.

I therefore doubt that a single replacement for seat pricing will dominate every category. Human access will remain a reasonable charging unit for work people continue to perform themselves. A base subscription with bounded agent usage may offer an acceptable compromise elsewhere.

This limits the prediction. Agents make an unexamined relationship between headcount and software value look outdated. They do not make every subscription a poor purchase.

A supplier can reasonably charge more for doing more work. The buyer can reasonably ask why it must retain the same access charges when the work has moved.

Watch the renewal, then the exception queue

British companies can test this argument at their next renewal without committing to a wholesale rebuild.

Choose a contained workflow with a recognisable completion point. Keep the existing route available during the trial. Compare the agent’s accepted results with the previous process, including the time people spend checking and correcting them.

Then examine what can actually change in the contract. Can licences be reduced during the term? Does automated access carry another charge? Will a spending cap stop the agent or merely trigger an alert? Ask who restores service when the agent fails during a busy period.

For the UK technology sector, I would watch two commercial signals closely. The first is whether customers reduce existing commitments after agents enter production. Selling an AI add-on demonstrates demand for AI; it does not establish that the old subscription has become redundant.

The second is whether suppliers accept responsibility for completed work, including exceptions. The opportunities described by Sage, Gradient Labs and Kainos depend on understanding or supporting the work around business records. That knowledge could command a price, but customers will need evidence that it produces dependable results.

My expectation is that recurring software revenue will endure, while the invoice becomes a mixture of access and work performed. The unresolved issue is how much of the productivity gain the buyer gets to keep.

An agent that completes a task with less human effort creates room for a better bargain. At renewal, British businesses need to claim their share of it.