Only 9% of B2B buyers consider per-seat pricing the fairest way to pay for AI agents, according to G2's 2026 AI Agents Insight Report, built on a July 2026 survey of 1,019 B2B decision-makers. Every alternative pricing model ranked higher — and 74% of buyers say they are open to replacing their current vendor at renewal.

The finding lands at a moment when agent spending has stopped being experimental. 86% of buyers say their company bought AI agent software or services in the past year, up from 69% in 2025. 66% have agents in limited or scaled production. At roughly half of companies in most industries, agent budgets now exceed $1M a year.

How fast did pricing preferences shift?

In one year, buyer preferences moved sharply toward metered models:

• Pay-as-you-go (consumption, per action or task): 38% in 2025 to 57% in 2026
• Pay-per-outcome: 11% to 19%
• Fixed pricing: 35% to 21%

The logic is structural, not sentimental. A seat counts humans who log in; an agent runs continuously and never logs in, so a per-seat price bears little relation to the work performed. Many B2B vendors added agents to existing per-seat plans in 2025 — a defensible first step, but one that now leaves them on the model buyers rank last heading into 2027 renewals.

Why are 74% of buyers shopping at renewal?

The renewal threat breaks down into measured segments: 30% of buyers would consider switching to a vendor with stronger agent capabilities outright, and another 44% would pilot an alternative and reconsider at renewal. 84% expect agent capabilities to influence vendor selection in 2027; 34% call it a primary deciding factor.

Three out of four customers, in other words, are willing to run a competitive pilot against the incumbent. Most B2B companies staff renewals with account managers; the report's implication is that large renewals in 2027 need to be staffed like new competitive deals.

Buyers are also specific about what "stronger" means: 72% say a prebuilt ecosystem of connectors, actions, skills, or marketplace offerings is very important or a must-have. Use of agent-builder platforms from Salesforce, ServiceNow, UiPath, Microsoft and others rose from 38% to 42%, while ready-made purpose-built agents fell from 30% to 27% — roughly doubling the gap between the two approaches, from about 8 points to 15. Thin APIs and MCP coverage now function as a sales blocker.

What happens when buyers get the meter they asked for?

Here the data cuts against vendors. 46% of agent buyers report higher-than-expected operating costs. Agents consume tokens, model calls, API calls and data services on their own schedule, and the bill arrives afterward.

Before accepting variable spend, buyers listed the controls they demand, in order:

1. Real-time usage dashboards
2. Monthly spend caps
3. Cost-per-outcome reporting
4. Routing to lower-cost models
5. Forecasting before deployment

Oversight compounds the cost problem. 31% of organizations say employees spend six or more hours a week supervising agents, rising to 40% at large enterprises. 45% say oversight and rework consume more than a quarter of the productivity gains they expected. A buyer paying per outcome who spends a day a week verifying those outcomes will negotiate the per-outcome price down hard.

Where is the $1M coming from?

Seven-figure agent line items are now common across industries — with one exception. The share of companies budgeting at least $1M a year for agents, 2025 to 2026:

• Healthcare: 47% to 57%
• Retail: 44% to 51%
• Financial services: 42% to 49%
• Technology: 41% to 48%
• Services: 40% to 47%
• Manufacturing: 39% to 34%

But the money is mostly new, not reallocated. Only 14% of companies report net headcount reduction from agents; 35% say they moved people to higher-value work. Most fund agents on top of existing headcount and software budgets. When CFOs begin demanding net savings — and they will — vendors with cost-per-outcome data will have an answer. Vendors reporting logins and seats will not.

What else does the report show?

• Sales development fell from 18% to 7% of agent-builder use cases in a year in G2's review data; process automation rose from 26% to 37% and software engineering from 13% to 23%. G2 notes this is a review sample from one category, not market share.
• 38% of organizations with agents in production run more than 100 of them; 47% still manage them informally or department by department.
• 77% re-evaluate production agents at least monthly.
• Central IT plus AI and data teams now own about 60% of agent governance accountability.
• 7% of buyers want fully autonomous agents; 48% want guardrails and 29% want them limited to specific tasks.

The report closes the loop on itself: budgets exist, preferences have moved, and the vendors that survive the 2027 renewal cycle will be the ones selling measured outcomes — with the dashboards and caps to prove them.