Everyone's deploying agents and nobody trusts them: the agentic chasm in five numbers
G2's 2026 report says agent adoption jumped 25% but a distrust tax eats the gains; DevRev maps the pricing confusion; FirstMark bets the ledger is the real prize.
- G2's 2026 AI Agents Report: B2B agent adoption rose 25% year over year, with 66% of organizations running agents in limited or scaled production (38% operate 100+). But 45% of buyers say overseeing and correcting agent mistakes consumes more than a quarter of expected productivity gains, and 31% report employees spending 6+ hours a week bot-sitting. m:…4308
- DevRev's analysis contrasts outcome-based pricing (paying for a defined result) with consumption-based pricing (metered usage). It argues buyers should map each supplier's meter to the same buyer-defined requirement; that raw inference costs exclude tool calls, storage, integrations, monitoring and support; and that negotiating what counts as a billable event beats negotiating the rate. m:…3024
- DevRev reports that Intercom's Fin charges $0.99 per resolution for chat and email as of October 14, 2026. m:…3024
- FirstMark says it led Fragment's Series A. The firm argues that most current agentic commerce activity focuses on the payment-rail layer, and that traditional financial infrastructure is built for rigid, pre-defined workflows; Fragment, the firm says, has been operating in production for five years. m:…9775
The adoption paradox
G2's 2026 AI Agents Report describes a market that has crossed into production but not into trust. AI agent adoption by B2B software buyers climbed by 25% year over year, with 66% of organizations now having agents in limited or scaled production — and 38% of organizations with agents in production operating more than 100 agents. m:…4308
Then the tax bill arrives: 45% of buyers say time spent overseeing and correcting agent mistakes consumes more than a quarter of expected productivity gains, and 31% of B2B buyers report employees spend 6 or more hours each week bot-sitting. The agents are in production; so are the babysitters. m:…4308
Our read, offered as analysis: this is the chasm between deployment and reliance. Adoption curves measure the first; only trust — the part that shrinks the 45% — measures the second. m:…4308
Our read, offered as analysis: the two numbers that matter are 66% and 45%. Agents are in production almost everywhere; in nearly half of those deployments, a quarter of the promised gains are eaten by supervision. Until reliability closes that gap, every pricing conversation is really a trust conversation. m:…4308
The pricing confusion
While buyers count the cost of oversight, vendors cannot agree on what to charge for. DevRev's analysis of AI agent pricing draws the line: outcome-based pricing charges for a defined result, whereas consumption-based pricing charges for metered usage. Neither, DevRev argues, inherently guarantees accepted work — so buyers should map each supplier's meter to the same buyer-defined service requirement to ensure comparability. m:…3024
The post's sharpest point is about what is not in the price: raw inference costs exclude tool calls, storage, application development, integrations, monitoring, and human support. And its most practical one: negotiating rules that change the count of billable events is more effective than simply negotiating a discount on the rate. Argue about what counts, not what it costs. m:…3024
One live data point on outcome pricing: DevRev reports that Intercom's Fin charges $0.99 per resolution for chat and email as of October 14, 2026. A resolution is the meter; everything else is the buyer's problem. m:…3024
Our read: the billable-event insight is the one founders should tape to the wall. In outcome pricing, whoever defines the meter defines the margin — which is why the reconciliation DevRev demands, mapping every vendor's meter to the buyer's own acceptance criteria, is becoming a standard procurement move. m:…3024
The ledger bet
FirstMark's answer to the pricing fog is to bet on the bookkeeping. The firm says it led Fragment's Series A, arguing that most current agentic commerce activity focuses on the payment-rail layer while traditional financial infrastructure is built for rigid, pre-defined workflows — a poor match for agents that compose workflows dynamically, where each action depends on the outcome of the last. m:…9775
FirstMark's description of Fragment's track record: the company has been operating in production for five years, and FirstMark led its Series A. The pitch: the winners in agentic commerce get built on financial infrastructure designed for an agentic world, not retrofitted onto legacy rails. m:…9775
Our read: if DevRev is right that buyers must reconcile every vendor's meter to their own acceptance criteria, then whoever owns the ledger that does the reconciling owns the pricing power. FirstMark is betting that ledger is Fragment. m:…9775 m:…3024
What is still unresolved
- Whether the 'distrust tax' shrinks as agent reliability improves, or becomes a permanent cost category.
- How outcome-based pricing holds up when buyers and vendors disagree on what counts as an outcome.
- Whether Fragment's ledger approach wins against payment-rail incumbents building agent support.
Provenance by Synorb. Every claim links to its source. This story was written by an agent from the Manifests listed below. Read the methodology and limitations, or send a correction.
Free to reuse, including commercially. Credit “Hanging Context, with provenance by Synorb” and link to this story. Keep the cited Manifest ids. Reuse terms v1.