Employees at 31% of companies spend 6+ hours a week reviewing agent output, fixing errors, and approving actions.
Key Takeaways
- In G2’s July 2026 survey of 1,019 business software buyers, 66% said their organizations use AI agents in live business work, beyond demos and pilots.
- 45% of respondents said that once they count the time spent overseeing and correcting agents, their expected productivity gains drop by more than 25%.
- 36% of respondents said their companies had agent-related security incidents in the past year.
- Among organizations with agents in production, 38% run more than 100 agents.
- Just 27% of respondents said their companies keep a central registry of their AI agents.
G2, a business-software review site, found that two-thirds of organizations now use AI agents in live business work. Nearly half of the buyers it surveyed said that when they factor in the time spent supervising those agents, their expected productivity gains drop by more than 25%.
The results come from G2’s 2026 AI Agents Report, based on a July 2026 survey of 1,019 business software buyers in North America, Europe, the Middle East and Africa, and Asia-Pacific.
Two-thirds of companies run agents beyond pilots
86% of respondents said their companies bought AI agent software or services in the past 12 months, up from 69% in G2’s 2025 survey.
66% said their organizations have agents in production. G2 counts an agent as in production when it works inside a live business process, using the company’s own data and systems, rather than in a demo or time-limited pilot. Production can be limited to a few teams, and it does not mean the agents act without human approval.
Among respondents whose agent pilots ended in the past year, 46% said more than half of those pilots went into full production.
At 45% of companies, supervising agents cuts expected gains by more than 25%
31% of respondents said employees spend six or more hours a week overseeing agent work. When asked which oversight task takes the most time, 26% of respondents named reviewing agent output, followed by 24% fixing agent errors, and 22% approving their actions.
45% said that once they count the time spent overseeing and correcting agents, their expected productivity gains drop by more than 25%.
Large companies report more oversight. At enterprises with 5,000 or more employees, 4 in 10 respondents said employees spend at least six hours a week overseeing agents, nearly twice the rate at companies with fewer than 250 employees. One in five said their expected gains drop by more than half once oversight and rework are counted, also nearly twice the rate at the smaller companies.
36% of companies had agent-related security incidents
36% of respondents said their companies had agent-related security incidents in the past year. G2 did not define an incident or specify what the incidents involved.
38% of companies with agents in production run more than 100, while just 27% keep a central registry
Among organizations with agents in production, 38% run more than 100 agents, and 15% run between 501 and 1,000.
47% of respondents said agent ownership and controls are informal or run mainly by individual departments. 23% said their agents are governed centrally.
Most companies also lack visibility into every agent deployed across their organization and what each one does:
- 27% of respondents keep a central registry of their AI agents.
- 18% employ automated tools that detect every agent, including ones the company never approved.
- Only 14% use real-time monitoring that flags unusual agent behavior, costs, or risks.
G2 compares untracked agents to shadow IT, employees’ use of software the IT department never approved. G2 says agents carry higher stakes than unapproved software because agents can access company data and start processes in company systems. When an agent behaves unexpectedly, G2 says the company needs to know who owns it, what systems and data it can reach, what actions it took, and how to recover.
Companies plan to give their agents more autonomy, but review them before and after launch
63% of respondents said they plan to let their current agents do more work on their own. 53% said they trust agents to act fully on their own in low-risk work.
Before an agent goes into production, respondents said their companies require these reviews:
- Human quality review: 52%
- Security review: 48%
- Legal, risk, or compliance review: 41%
After launch, 77% recheck their agents’ performance at least once a month.
By 2028, 54% of respondents expect agents to manage more than a quarter of their organization’s workflows, and 25% expect them to manage more than half.
Methodology
G2 surveyed 1,019 business software buyers who make or influence software purchases for departments, operating units, or whole companies. Their titles ranged from individual contributor to vice president and above. G2 fielded the online survey in July 2026, with respondents in North America, Europe, the Middle East and Africa, and Asia-Pacific.
G2 also drew on more than 2,500 reviews on its site from March to August 2026 to see how companies use agents. It said it used generative AI reasoning models to help design the survey and analyze the results.
The hours and lost gains are respondents’ own estimates, not measured figures.
