By PYMNTS | October 22, 2026
Imagine hiring a new employee who works around the clock, never takes a break and can place orders or move money on its own. Now imagine hiring that worker without checking references. That is roughly the risk companies take when they rush artificial intelligence agents into their operations.
AI agents are tools that can plan and carry out tasks with little human input. Companies need a smarter way to try these tools before rolling them out, attorneys Spencer Glende and Samuel J. Hartman of law firm Mayer Brown said in a Thursday (Oct. 22) article that appeared on Mondaq. The answer is a well-designed pilot program backed by a strong contract.
Companies tend to fall into one of two traps, the article said. Some move too fast and launch agents before they understand what the tools can and can’t do. That can lead to wasted money, legal trouble or damage to a company’s reputation. Others get so worried about the risks that they give up on tools that could deliver real value.
The stakes are high either way, according to the article. Depending on how they are used and controlled, “AI agents may bring game-changing gains or cause irreversible harm.”
The authors said their advice is to treat a pilot like a job interview. A careful employer checks whether a candidate has the right skills, can be trusted and is worth the salary. Companies should ask the same three things about an AI agent. Can it do the job? Is it safe? Will it pay off?
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
That starts with a clear job description, the article said. The contract should spell out which tasks the agent will handle, what data and systems it can reach and who can work with it. It should also set limits. A purchasing agent, for example, might be allowed to draft orders under a certain dollar amount but need a human to approve them.
Next comes testing, according to the article. Companies should track how often the agent finishes tasks, how accurate it is and how often it makes things up or oversteps its limits. Tests should include tricky situations, confusing requests and attempts by fraudsters to fool the system. The agent should also keep detailed records of what it did and why.
Safety checks need to go beyond rules on paper, the article said. A company should confirm that the agent actually respects spending caps and privacy rules in practice. It should also test how fast an emergency off switch works. Human reviewers need a real chance to catch problems. Too many alerts can backfire, since busy staff may start ignoring them.
Contracts can only do so much, according to the article. A contract can decide who pays when something goes wrong. It can’t stop the failure from happening in the first place. That is why hands-on testing and careful vetting of the vendor still count.
Cost is the third test, the article said. Fees for the software are only part of the bill. Computing charges and the extra staff needed to handle problems can add up quickly. An agent might replace some workers but require more supervisors.
So, what should companies expect next? Every pilot needs a clear end date and a firm decision point, according to the article. A pilot should never slide into full use just because no one stopped it. The contract should also cover what happens if the test falls short. Options might include a longer trial, more fixes or a smaller rollout instead of a simple yes or no.
Done right, a pilot helps companies avoid both rushing ahead and walking away too soon, the article said.
For all PYMNTS AI coverage, subscribe to the daily AI newsletter.