The Distrust Tax on AI Agents Is Costing Six Hours a Week

AI Agents Are Saving Hours across the modern workplace, or at least that was the promise. New research from the software marketplace G2 suggests the savings are leaking away: nearly a third of companies spend six or more hours each week overseeing their agents' work, and almost half say checking and rework consume more than a quarter of the productivity gains they expected. The company has a name for the gap, calling it the Distrust Tax in its latest agents report.

The findings land at an awkward moment for the agent hype cycle. Adoption is no longer the question. Almost nine in ten B2B software buyers purchased AI agent software or services in the past year, a jump of a quarter from the year before, according to the PR Newswire release on the study. Two-thirds of buyers now have agents in limited or scaled production, up nearly a fifth year over year. The agents are hired. The problem is that nobody quite trusts them to work alone.

Where the missing hours go

Companies are deploying AI Agents faster than they can govern them, and the oversight bill shows up in three places: time, money, and governance. On time, the report says one-third of organizations with agents in production spend at least six hours a week on what G2 calls bot-sitting, the labor of watching agents work and fixing what they get wrong. That is nearly a full workday each week spent supervising software that was purchased to reduce work, a paradox that should make any manager wince.

On money, the meter is running faster than buyers expected. Nearly half of agent buyers report higher operating costs than they planned for, according to coverage of the report by the SaaS outlet. Agents burn tokens, model calls, API calls, and data services on their own schedule, and the bill arrives after the fact. Half of organizations now put their annual agent budgets at a million dollars or more, which means the oversight labor sits on top of a serious line item, not a side experiment.

On governance, the controls have not caught up with the deployment. Roughly half of organizations manage agent ownership and controls informally or through individual departments. Just over a quarter maintain a central registry of their agents, and fewer than one in seven run tools that monitor unusual behavior, costs, or risks as they occur. Nearly two in five organizations run a hundred or more agents, so the informal approach is being asked to cover real scale. It is the corporate equivalent of hiring a hundred interns and keeping track of them on sticky notes.

The tax is real, but so is the payoff

None of this means the agents are useless, and the report is careful to say so. More than four in five buyers say agents improved employee satisfaction by reducing repetitive work. A third have moved staff into higher-value roles, and only a small share report net headcount cuts, according to the HRD Canada report on the findings. The productivity story is not a myth; it is being discounted by the cost of verifying it.

Tim Sanders, G2's chief innovation officer, put the tension plainly: agents are showing real signs of value, but much of it is being erased by oversight, rework, and exception handling, according to the industry coverage. Sanders argues the fix is not fewer agents but better scaffolding: clearer evidence of how agents behave inside company systems, clear ownership of each one, and cost controls that catch runaway spend before the invoice lands. For agents to reach mass adoption, in other words, companies have to stop treating them like magic and start managing them like employees.

That shift has a direct message for young workers entering the job market now. The scarce skill is no longer doing the task the agent does; it is supervising the agent, spotting its mistakes, and knowing when to override it. Tech postings hit a three-year high even as hiring shrank, and the workers who thrive in that split-screen market may be the ones who can manage a fleet of tireless but untrustworthy digital interns. Judgment, not output, is becoming the premium.

The market seems to agree that trust is the bottleneck. More than five in six buyers say agent capabilities will influence their 2027 software choices, with about a third calling them a primary deciding factor, according to the release. That is a lot of purchasing power hinging on whether vendors can prove their AI Agents behave. Until then, the Distrust Tax keeps collecting, one checked hour at a time. Workers curious about how the other half of the labor market is adapting can also read GenZ NewZ's coverage of the record independent workforce and the Gen Z job search wave.

Sources: the press release, AI Brief, the Canadian workplace report.