Most SaaS pricing pages still answer one question: how many people need a login? That question is losing its grip on AI agents, because an agent doesn't log in. It runs all night, calls other tools, burns tokens, and finishes work nobody watched it do.
Buyers have noticed. The G2 2026 AI Agents Report, published October 14, 2026 from a July survey of 1,019 B2B software decision-makers, finds only 9% of buyers consider per-user or per-seat pricing the fairest model for agents. Preference for pay-as-you-go models (consumption plus per-completed-task billing) rose from 38% in 2025 to 57% in 2026.
This B2Bcentr report shows which sample produced each number, how far vendor pricing has already moved, why variable pricing is scaring the same buyers who asked for it, and what to change on a pricing page this quarter. It's the pricing companion to B2Bcentr's 2026 B2B GTM AI impact report, which tracked how AI is changing the selling side.
Key Takeaways
- Only 9% of buyers call per-seat fair for AI agents.
- Pay-as-you-go preference rose from 38% to 57%.
- 46% say agent operating costs ran higher than expected.
- 95% of AI agent vendors already price hybrid.
- 74% would shop a rival with stronger agents at renewal.
Which sample produced the 9%
Primary source: the G2 2026 AI Agents Report, subtitled Crossing the Agentic Chasm. G2 fielded an online survey of 1,019 B2B decision-makers who make or influence software purchases, in July 2026, across North America, EMEA, and APAC. Company sizes run from SMB (1 to 249 employees) to large enterprise (5,000+). G2 also referenced over 2,500 reviews in its AI Agent Builder category from March to August 2026.
The exact question matters. G2 asked which pricing model buyers consider fairest for agent services. That's a stated preference, not a count of signed contracts. It tells a vendor what the buyer will argue for in a negotiation, which is still worth knowing before the renewal call.
| Buyer pricing preference for AI agents | 2025 | 2026 | Change |
|---|---|---|---|
| Pay-as-you-go (consumption or per completed action/task) | 38% | 57% | +19 pts |
| Pay-per-outcome | 11% | 19% | +8 pts |
| Fixed pricing | 35% | 21% | -14 pts |
| Per-user or per-seat called fairest | n/a | 9% | 2026 only |
Source: G2 2026 AI Agents Report (N=1,019 B2B buyers, surveyed July 2026). Point changes are B2Bcentr arithmetic on G2's published shares.
The folklore: price the agent like another seat
The comfortable SaaS playbook says AI is a feature. Bundle it into a higher tier, or sell it as a per-seat add-on, and keep the ARR model intact. Plenty of CRMs still price this way, which B2Bcentr's CRM cost report shows SKU by SKU.
G2's buyers reject the logic for agents specifically. In the report's pricing chapter, G2 argues that an agent draws on models, data, APIs, and other software as it works, so human logins have become a weak proxy for the value it creates. Fixed pricing lost 14 points of preference in a single year, and outcome pricing gained 8.
Vendors have already moved further than buyers think
The supply side isn't standing still. The Orb 2026 AI agent pricing study of 80 AI agent companies found 95% use hybrid pricing, up from 92.4% a year earlier. 91.3% meter and charge for usage, up from 83.3%, while 71.3% still carry a subscription component.
Here's the gap worth noticing. Outcome-based pricing sits at just 3.8% of vendors in the Orb sample, down from 4.5% in 2025, while 19% of buyers in the G2 survey now prefer pay-per-outcome. That's roughly a 5x gap between what buyers say they want and what vendors actually sell (B2Bcentr arithmetic across two different samples, so treat it as direction, not precision). Orb attributes the shortfall to instrumentation: defining and attributing an outcome is hard, and most billing systems weren't built for it.
| Pricing model | Vendor share (Orb, 80 cos) | Buyer preference (G2, 1,019 buyers) | Published example |
|---|---|---|---|
| Hybrid (subscription plus usage) | 95% | not asked as hybrid | The default; Orb calls it the baseline |
| Usage or consumption | 91.3% meter usage | 57% pay-as-you-go | Credits, tokens, per-action fees |
| Subscription component | 71.3% | 21% fixed | Platform fee or tier |
| Outcome-based | 3.8% | 19% | Intercom $0.99 per resolution; Gorgias $0.90 per resolved interaction |
Sources: Orb 2026 AI agent pricing study (80 AI agent companies; vendor list prices as cited by Orb) and the G2 2026 AI Agents Report. The two samples measure different things: vendor practice versus buyer preference.
Customer support is where outcome pricing already has a public rate card. Per the Orb roundup of published rates, Intercom lists $0.99 for a resolution outcome and Gorgias lists $0.90 per resolved interaction on most plans, with Starter plans beginning at $1. Support works because a resolved ticket is easy to count. Most B2B agents (research, SDR, ops) don't have an outcome that clean yet.
The catch: 46% say agents cost more than planned
Buyers asked for variable pricing, then got the variable bill. In the G2 2026 AI Agents Report, 46% of agent buyers report higher-than-expected operating costs. 31% say employees spend 6 or more hours a week supervising agent work, and 45% say oversight and rework eat more than a quarter of the productivity gains they expected. G2 calls this the Distrust Tax.
So the buyer who wants pay-as-you-go also wants guardrails on it. G2 lists the top controls buyers want before they'll accept rising variable costs, in this order:
- Real-time usage dashboards
- Monthly spend caps
- Cost-per-outcome reporting
- Lower-cost model routing
- Forecasting before deployment
G2's own line in the pricing chapter sums it up: "Before buyers will accept rising costs, they expect the controls that make spend visible and manageable." A meter without a cap reads as a blank check.
That warning comes straight from Aleph's 2026 gross margin benchmark: usage-only models post a 62% median total gross margin, well below the 76% to 84% range for subscription variants, because consumption revenue carries infrastructure and compute cost. B2Bcentr's SaaS gross margin report covers the full cut, and the SaaS NRR report shows the retention upside usage models tend to earn in exchange.
Renewal is where seat pricing gets tested
The pricing argument now arrives with a switching threat attached. In the G2 survey, 30% of buyers would consider switching to a vendor with stronger agent capabilities, and another 44% would pilot an alternative and reconsider at renewal, a combined 74%. 84% expect agent capabilities to influence their 2027 software choices, with 34% calling it a primary deciding factor. 72% rate a strong ecosystem of connectors, actions, skills, or marketplace offerings as very important or a must-have.
Adoption explains the urgency. The same report finds AI agent purchasing among B2B software buyers rose from 69% to 86% year over year, and 66% of organizations now run agents in limited or scaled production. 38% of organizations with agents in production operate more than 100 of them. Per-seat math breaks fast when one buyer runs 100+ agents and five humans.
Which model fits which agent
| Agent type | Best-fit unit | Main risk | Control to ship first |
|---|---|---|---|
| Customer support | Per resolution (Intercom lists $0.99) | Disputes over what counts as resolved | Outcome definition in the contract |
| Research or enrichment | Per record or per credit | Runaway token spend | Monthly spend cap |
| SDR or outbound | Per qualified meeting or per action | Paying for low-quality meetings | Cost-per-outcome reporting |
| Internal ops or workflow | Platform fee plus usage tier | Bill shock on big batch runs | Forecast before deployment |
| Agent builder platform | Hybrid: seats for builders, usage for runs | Mixed signals at renewal | Real-time usage dashboard |
B2Bcentr synthesis. Controls ranked by buyers in the G2 2026 AI Agents Report; vendor model shares and the Intercom rate from the Orb 2026 AI agent pricing study.
The platform row matters more each quarter. The G2 survey found buyers using agent-builder platforms rose to 42% in 2026 while ready-made agents fell to 27%. Builders pick platforms for control, which means they'll scrutinize the meter just as closely.
Honest limits of the 9%
- It's stated preference. Fairest isn't the same as what buyers will sign, and seat-based contracts still renew every day.
- G2 sells marketing and research services to software vendors, so the report has a point of view on what vendors should do next.
- G2 says generative AI models helped define the focus areas, design the survey, and analyze results. The sample size is solid, but the framing is G2's.
- The vendor shares come from a different sample (80 AI agent companies in Orb's study), so the buyer-versus-vendor gap is directional.
- No public 2026 number yet shows that switching off seats raises win rates or NRR for agent products specifically.
What to change on a pricing page this quarter
- Name the unit. If an agent does the work, price the work: a resolution, a record, a meeting, a run.
- Keep a floor. A platform or subscription fee protects gross margin while usage ramps.
- Publish the cap. Show a monthly spend limit and an overage rule on the page, not only in the order form.
- Report cost per outcome. Put it in the product, so finance can defend the renewal.
- Pilot outcome pricing on one agent where the outcome is countable before rolling it out.
FAQ
How do AI agent companies price their products in 2026?
Most use hybrid pricing. 95% of 80 AI agent companies in the Orb 2026 AI agent pricing study combine a subscription with usage charges, and only 3.8% price on outcomes.
Is per-seat pricing dead for AI agents?
Not dead, but losing the argument. Only 9% of buyers in the G2 2026 AI Agents Report call per-seat the fairest model for agents, so expect pushback at renewal.
What pricing model do B2B buyers prefer for AI agents?
Pay-as-you-go. 57% of buyers prefer consumption or per-completed-task pricing in 2026, up from 38% in 2025, per the G2 2026 AI Agents Report.
Why are AI agent costs higher than expected?
Because agents consume compute and need supervision. 46% of buyers report higher-than-expected operating costs, and 31% spend 6 or more hours a week overseeing agents (G2).
How much does outcome-based AI agent pricing cost?
Published support rates sit near $1 per resolution. Intercom lists $0.99 per resolution and Gorgias $0.90 per resolved interaction, as cited in the Orb 2026 AI agent pricing study.
Does usage-based pricing hurt SaaS gross margin?
Yes, at the median. Aleph's benchmark puts usage-only models at a 62% gross margin versus 76% to 84% for subscription variants.
What controls do buyers want before accepting usage pricing?
Dashboards and caps first. The G2 2026 AI Agents Report ranks real-time usage dashboards, monthly spend caps, cost-per-outcome reporting, cheaper model routing, and pre-deployment forecasting.
Next read: pricing model changes show up in retention first. See how usage-based and seat-based companies compare in the B2Bcentr NRR report.
Pricing pages tend to lag the market by a year. Once the meter is set, the next number to watch is retention, which B2Bcentr's 2026 SaaS NRR report breaks out by pricing model.