Your Team Spends Six Hours a Week Bot-Sitting. G2 Calls That the Distrust Tax.

Opening & At a Glance

It's 3:40 on a Thursday. Someone in ops is still reviewing the ticket queue an agent was supposed to clear by lunch. The summaries look fine until they don't — a wrong account, a skipped approval, a polite reply to a customer who needed a human. By the time the corrections are done, the afternoon is gone. The agent worked. So did the person hired to watch it.

That scene is what G2's new research is trying to name. On 14 October 2026 the company published its 2026 AI Agents Report: Crossing the Agentic Chasm, based on a survey of more than 1,000 B2B buyers. The headline isn't that agents failed. It's that they made it into production, and the cost of not fully trusting them followed them there.

What G2 Found, Without the Brochure Voice

Last year's G2 Agents Report coined the Trust Gap — buyers liked agents but wouldn't let them run free. This year the gap didn't close when agents entered real workflows. It moved inside them. G2 calls the resulting cost the Distrust Tax: reviews, corrections, exception handling and plain bot-sitting that claw back the speed the pitch deck promised.

The production numbers look strong on paper. Two-thirds of surveyed agent buyers have agents in limited or scaled production. Nearly half of organisations that finished pilots say most of those pilots went live. Four in five teams running agents in mature, high-volume workflows report cycle-time gains of at least 11%. That's real. It's also the part vendors will quote.

Then the tax. Almost a third of buyers say staff spend six or more hours a week overseeing agent work. Forty-five percent say oversight and rework consume more than a quarter of the productivity gains they expected. Forty-six percent say operating costs came in higher than planned. Large enterprises feel it hardest in G2's cut: 27% say they need significantly more oversight, nearly double the 15% SMB rate.

Scale makes the mess worse. Thirty-eight percent of organisations with agents in production run more than 100 of them. Almost half still manage ownership informally or department-by-department. Only 27% keep a central agent registry; 14% claim real-time observability into odd behaviour, cost spikes or risk. G2's phrase is blunt: shadow agents are the new shadow IT.

Confirmed vs Analysis — Keep the Labels Honest

Why the Distrust Tax Matters More Than Another Pilot

Here's my read. The market spent two years selling autonomy. Buyers bought capability. What they got, in many shops, is a second job: watching the first one. Six hours a week of bot-sitting isn't a footnote. It's almost a full day of skilled labour spent verifying work that was supposed to free people up. If you're measuring "agent ROI" without subtracting review time, you're measuring a fantasy.

This is the buyer-side twin of what vendors have been shipping. Last week we wrote about Cohere North 2's token caps and spend alerts — brakes as a feature. The week before that, OpenClaw's enterprise control plane made the same bet from the open-source side: inventory, ownership, kill switches. G2's survey is the demand curve those products are answering. Buyers aren't asking for a smarter demo. They're asking who owns the agent, what it costs when it loops, and how fast a human can stop it.

G2 leans on Geoffrey Moore's chasm metaphor, and for once it fits. Visionaries funded pilots with AI war chests. Pragmatists want proof that holds up after oversight. Fifty-two percent of surveyed buyers require human quality review before production. Seventy-seven percent re-evaluate performance at least monthly. Only 7% say they're comfortable with full autonomy. That isn't fear of AI. It's procurement doing its job.

One more signal: just 9% of buyers think per-seat pricing is fairest for agents. Pay-as-you-go jumped hard in G2's year-over-year cut. Seat pricing was built for humans who sit still. Agents don't. If your SaaS renewal still assumes a headcount of seats while agents do the work, expect a fight in 2027 — G2 says 84% of buyers already expect agent capability to shape next year's software choices.

Questions Readers Are Likely to Ask

Is "66% in production" universal? No. It's from G2's survey of 1,019 B2B decision-makers who influence agent purchases. A buyer-side signal, not a census.

Trust Gap vs Distrust Tax? G2's 2025 framing was hesitation before deployment. The 2026 framing is the ongoing cost — time and money — of running agents you still don't fully trust once they're live.

Does G2 say agents aren't worth it? No. The same report cites cycle-time gains among respondents. The argument is that net value depends on cutting oversight cost, not on abandoning agents.

Closing & Sources

Confirmed: G2 published the 2026 AI Agents Report on 14 October, from a July survey of 1,019 B2B buyers, and the Distrust Tax stats above come from that survey — findings, not universal truth. Opinion: the chasm isn't "will agents work?" anymore. It's "can we afford the humans who still have to babysit them?" The vendors racing to sell registries, caps, kill switches and proof packs aren't late to the party. They're finally selling what the pragmatists asked for.

• G2 company blog — "New G2 Research: AI Agents Have Moved From Pilot to Production. Now Comes the Distrust Tax." (Tim Sanders, 14 Oct 2026)
• PR Newswire — "New G2 Research: AI Agents Have Moved From Pilot to Production, But Oversight Comes at a Cost" (14 Oct 2026; methodology and expanded stats)
• G2 Insights — 2026 AI Agents Insight Report landing page (Crossing the Agentic Chasm)

Methodology Note: Editorial synthesis of G2's 14 Oct 2026 company post and the matching PR Newswire release. All percentage figures are G2 survey findings from 1,019 B2B decision-makers (July 2026) unless labelled otherwise. RogueGhost is not affiliated with G2. Analysis labelled as analysis.