AI agents are quickly moving from experiments into the workplace, but a new survey suggests many companies may be moving faster than their ability to control them.
Finance departments are under pressure to deploy AI as quickly as possible, even as questions around security, accountability, and oversight remain unanswered.
That is the uncomfortable takeaway from new research released by Avalara. The company surveyed more than 1,500 CFOs and senior finance leaders across the United States, United Kingdom, India, and Australia who had deployed, tested, or evaluated AI agents in financial workflows over the past year.
The message from executives is clear: they want AI, and they want it now. According to the survey, 92 percent of finance leaders said they feel career pressure to prove their AI investments are delivering results. Half called that pressure significant.
The problem is that many organizations appear to be prioritizing deployment speed over making sure they are ready.
Seventy-one percent of respondents said the pressure they face is mostly about how quickly they can deploy AI agents. Only 7 percent said their organizations put governance ahead of speed. That is a concerning gap.
Traditional software usually does what it is programmed to do. AI agents are different. They can analyze information, make recommendations, and in some cases take actions on behalf of employees.
That creates a new challenge: when an AI agent makes a bad decision, who is responsible?
The survey found that nearly one quarter of finance leaders believe accountability for a major AI agent failure would either be unclear or belong to nobody. Others pointed toward the employee managing the agent, the team that deployed it, or the executive who approved the investment.
That is not exactly a comforting answer for a technology being introduced into areas like tax, compliance, and financial reporting.
Another issue is that many companies do not have enough internal knowledge about how these systems work. Seventy-six percent of respondents said they lack dedicated in-house expertise to understand their AI agents.
In other words, some organizations are deploying AI systems that they may not fully understand.
The governance problem goes beyond staffing. Thirty percent of respondents said their companies have not updated internal controls within the last year to account for AI agents taking or recommending actions. Nearly half said their AI incident response plans are either still being developed or have not been tested.
Finance leaders are not rejecting AI. They simply want more confidence before giving these systems more responsibility.
The survey found that executives want AI agents that operate within existing business systems, use verified financial and compliance data, and provide documentation showing what the agent did and why.
That seems reasonable. A company does not just need an AI agent that can act. It needs one that can explain itself afterward.
The AI industry has spent the last few years showing off what these systems can do. Now businesses are entering the harder phase: figuring out how to safely put them in charge of real work.
Moving fast with AI may create an advantage, but moving too fast could create a different kind of problem. If an autonomous system makes an expensive mistake, “the AI did it” probably will not satisfy regulators, customers, or shareholders. The companies adopting these tools will need clear answers before those questions arrive.
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz