What changed
G2 released its 2026 AI Agents Report on October 14, based on an online July survey of 1,019 B2B software purchasing decision-makers in North America, EMEA and APAC. Only 9% rated per-user or per-seat pricing fairest for agent services. Pay-as-you-go preferences, including consumption and completed tasks/actions, reached 57% versus 38% in its 2025 comparison; pay-per-outcome rose from 11% to 19%, while fixed pricing fell from 35% to 21%. Meanwhile 46% reported higher-than-expected agent operating costs, 31% at least six hours a week supervising agents, and 45% said supervision and corrections erased more than one-quarter of expected productivity gains.
Why it matters
Small SaaS and AI-agent vendors cannot assume an existing seat-based subscription is an intuitive value meter for agents that consume models, APIs and compute around the clock. But a usage meter alone can introduce unbounded spend anxiety. G2's responses suggest buyers want real-time dashboards, monthly caps, cost-per-outcome reports, lower-cost routing and pre-deployment forecasts. Builders should test pricing against measured customer workflows and fully loaded cost, not infer that all buyers will switch simply because a survey preference changed.
Seat count is a poor proxy for agent work
In G2's 2026 sample, only 9% selected seats as the fairest pricing basis for agents. Pay-as-you-go reached 57% and pay-per-outcome 19%, compared with 38% and 11% in the report's 2025 comparison. These are stated preferences for agent services, not proof that 57% of agent purchases already use consumption billing, nor evidence that conventional SaaS seats have disappeared.
The bill includes supervision, not just tokens
Forty-six percent of respondents reported higher agent operating costs than expected. Thirty-one percent said employees spend six or more hours weekly supervising agents; 45% said rework and oversight consume more than a quarter of expected productivity benefits. That points to a total-cost model incorporating model calls, data access, review, exception handling and recovery rather than tokens alone.
Spend controls are part of the product
G2 says buyers prioritised real-time usage dashboards, monthly spending caps, cost-per-outcome reporting, lower-cost model routing and cost forecasting. A small vendor adopting usage pricing should define the billable event, expose meter semantics, let customers budget and cap consumption, and provide dispute and retry handling before making the meter central to a contract.
Renewal interest is not churn
The report says 30% would consider switching for stronger agent capabilities and another 44% would pilot alternatives at renewal. These are intentions, not observed cancellations. G2 also reports 72% value a strong ecosystem of connectors, actions or skills; this favours usable integrations and measurable workflow outcomes over adding a nominal AI feature.
Read the methodology before treating the percentages as market shares
G2 surveyed 1,019 people who influence or make B2B software purchasing decisions in July 2026, across several regions and company sizes. It also drew on more than 2,500 reviews in its Agent Builder category for use-case context. The sample is not a random census of all businesses, 2025-to-2026 comparability is not independently audited here, and self-reported spending or productivity changes are not independently measured financial records.