Enterprises boost AI agent spend but retain human checks

Most large companies plan to increase spending on AI agents next year, as early deployments spread across core business functions and security concerns hold back full automation.

A new survey commissioned by automation platform Zapier found that 84% of enterprise leaders expect to boost investment in AI agents over the next 12 months. The study focused on senior executives at US companies with more than 1,000 employees.

The research showed that 72% of enterprises already use or test AI agents. Within that group, 40% run multiple agents in production.

The findings indicate that so‑called agentic AI is moving into mainstream business planning. Many respondents see it as a key element of 2026 strategy.

The survey collected 525 responses from C-suite executives, presidents, owners and partners. Research firm Centiment conducted the fieldwork.

Zapier describes AI agents as tools that can perform tasks autonomously across different software systems. Many companies deploy them inside existing workflows.

"Enterprises are rushing to adopt AI agents because the potential for automating routine workflows is massive; 30% of leaders see this as the greatest potential use case," said Emily Mabie, AI Automation Engineer, Zapier. "However, the enthusiasm is tempered by some skepticism. Our data shows that while AI agents can operate autonomously, most companies still prefer a 'human-in-the-loop' approach."

The results point to a gap between the appetite for automation and the level of trust in fully independent systems. Many enterprises keep people involved at key decision points.

Spending and uptake

Investment plans were strong across the sample. The survey reported that 84% of leaders said it was likely or certain that their organisation would increase AI agent spending within a year.

Adoption has already spread beyond pilots in many cases. Almost three-quarters of respondents said their companies were either using or testing AI agents.

Forty per cent of those with deployments said they had more than one agent in production. This suggests that some organisations now run portfolios of agents across functions.

The data also indicated early clustering in particular departments. Customer support and operations ranked as the top areas for agent deployment.

Zapier reported that 49% of enterprises had introduced AI agents into customer support. It said 47% had done so in operations.

Security concerns

Security and data privacy emerged as the leading barriers to wider AI agent adoption. Respondents cited risk considerations as a key reason for caution.

Zapier said these worries echoed findings from an earlier study on AI tool sprawl. In that research, 36% of enterprises reported higher security and privacy risks from disconnected AI tools.

The new survey suggested that companies want more control and stronger governance around agents that can act across multiple systems.

Many executives appear reluctant to grant agents broad autonomy without clear safeguards. They expressed concerns about data access, compliance and oversight.

Use cases and workflows

The study highlighted data management as the most common use case for AI agents. Nearly half of enterprises, 47%, said they used agents for tasks such as data entry and extraction.

These tasks often involve high volumes of repetitive work and structured information. Many organisations already use software for such activities, which makes them a natural target for automation projects.

Survey respondents also pointed to routine workflow automation as a major area of interest. Many leaders identified these processes as a source of potential efficiency gains.

Mabie said many businesses now view AI agents as part of broader process orchestration efforts. She noted that this involves coordination between tools rather than isolated deployments.

"Companies need to shift their focus from simply acquiring the latest AI tools to properly orchestrating them into unified business processes," Mabie added. "An integrated approach is the only way to ensure these new AI agents enhance productivity, rather than creating new organizational headaches and costs."

Zapier said enterprises were experimenting with no-code systems that let staff design and deploy agents themselves. It said these systems could reduce reliance on central IT teams while maintaining governance controls.

The company reported that most organisations in the survey still favoured a human-in-the-loop model. It said respondents often inserted checkpoints into workflows where employees review or confirm actions proposed by agents.

Zapier said it expects enterprise interest in AI agents to rise as businesses refine security standards and integration practices over the next year.