G2's 2026 AI Agents Report: The "Distrust Tax" — Oversight, Not Capability, Is Now the Adoption Bottleneck
G2 published its 2026 AI Agents Report, "Crossing the Agentic Chasm," on October 14, 2026 (online survey of 1,019 B2B decision-makers responsible for or influencing purchases). The headline for anyone selling agentic software: agents have crossed into production, and the bottleneck has moved from capability to trust-driven overhead — which G2 names the "Distrust Tax." This is the sequel to G2's 2025 Agents Report "Trust Gap" finding, and it hardens into concrete evaluation criteria.
Adoption is real and compounding. 86% of B2B software buyers purchased AI agent software or services in the past year (up from 69% a year earlier), and 66% now run agents in limited or scaled production (+18pp YoY). 38% of organizations run 100+ agents, and half have an agent budget of at least $1M (healthcare leads spend: 57% at $1M+).
But the gains are being taxed by oversight. Per the release: 31% of buyers say employees spend 6+ hours per week "bot-sitting" agent work; 45% say oversight and rework consume more than a quarter of the productivity gains they expected; 46% report higher-than-expected agent operating costs. Enterprises pay the highest tax — 27% require significantly more oversight, nearly double the SMB rate of 15%.
"AI agents are showing real signs of value, but much of it is being erased by oversight, rework, and exception handling… For agents to cross that chasm, pragmatists need real proof of credibility — not more marketing promises, but repeatable results in real workflows and the controls that make autonomy safe to extend." — Tim Sanders, Chief Innovation Officer, G2 (PR Newswire, Oct 14, 2026)
A governance vacuum sits behind the tax. 47% of organizations still manage agents informally or department-led (vs. 23% centrally governed); only 27% maintain a central AI agent registry; just 14% have real-time observability into unusual agent behavior, costs, or risks. This is the buyer-side gap that frameworks like Agentic-Washing Meets the Governability Gap: RFPs Now Score Policy Enforcement, Audit Trails, and Cost Predictability predict RFPs will score.
The new evaluation criteria for agent vendors (what buyers now demand):
- A higher proof bar than traditional software: half of recent agent buyers use their own data science teams to verify vendor claims; before production, 52% require human quality review, 48% security testing, 41% legal/risk/compliance review. After deployment, 77% re-evaluate agents at least monthly — the renewal is effectively continuous.
- An autonomy ceiling: 48% prefer agents with guardrails, 29% want them limited to specific tasks, and only 7% are comfortable with full autonomy.
- Pricing rejection of seats: just 9% consider per-user/per-seat the fairest model for AI agents; pay-as-you-go (consumption/completed-task) preference rose from 38% in 2025 to 57% in 2026 — reinforcing The Shift from Seat-Based to Outcome-Based AI SaaS Pricing: Vendor Playbooks and Procurement Realities.
- Ecosystem as a gate: 72% say a strong connector/action/marketplace ecosystem is very important or a must-have; 84% say agent capabilities will influence their 2027 software choices (34%: primary deciding factor); 56% expect agents to orchestrate work across their SaaS stack by 2030.
What a founder selling to enterprises should take from this: sell the controls, not the autonomy — guardrails, audit trails, registries, and observability are what shorten deals; assume the customer will verify claims with their own data scientists and re-evaluate monthly; price for consumption or outcomes because seats have lost legitimacy for agent workloads; and treat your connector ecosystem as a procurement checkbox, not a developer nicety. The buyer-side mirror of this report — agents used in the buying process itself — is covered in G2's "Evaluation Maze": Evaluation Is Now the Longest Buying Stage — and the CFO Is Vetoing Approved Deals and G2's Answer Economy: AI Chatbots Are Now the #1 Shortlist Gate, and Review Citations Are the Trust Signal.