By Belle Lin
With the adoption of artificial-intelligence agents becoming more widespread among businesses, some of the most avid corporate users are racing to avoid the issues that come with having too many agents.
Companies including Lyft, DaVita and GitLab are navigating the challenge of AI agent proliferation -- hoping to tamp down on the problem without discouraging AI use, they told The Wall Street Journal.
The issue, known as "AI agent sprawl," stems partly from how easy it is for even nontechnical employees to create these independent AI bots, thanks to platforms like Anthropic's Claude Cowork. OpenClaw, an open-source tool that orchestrates multiple agents, has also added to the popularity of agents at work.
But having too many AI agents, especially multiple agents performing the same tasks, can create a cybersecurity and management problem for corporate information-technology departments, and can drive up computing bills, those companies said.
"Because everybody can do it, we're probably going to end up with a lot of people having the same types of agents," said Michael Friedlander, chief information officer of the Americas at Ben & Jerry's owner, Magnum Ice Cream.
The ice-cream maker will eventually need to condense and centralize all of the AI agents created inside the company, though it's aiming not to disrupt employee creativity in the process, Friedlander said.
"Depending on how all of this will turn out, there'll be tokens, and then there's going to be cost, and then you end up with, 'How do we manage this to make sure that it's under a financially responsible model?'" he added.
In the next two years, the average global Fortune 500 enterprise will run over 150,000 AI agents, according to market research and IT consulting firm Gartner. But only 13% of organizations think they have adequate AI agent governance in place, Gartner said.
Managing the risks of sprawl
For now, most AI agents are being deployed by white-collar workers to do things like write code, summarize emails, perform customer support and analyze data. But more complex agents are being used for "deep research" and automating entire business workflows.
Even so, they could be running on an employee's laptop, a server or other company systems, making it hard for IT departments to keep track of them all, said Jai Das, co-founder and president of venture-capital firm Sapphire Ventures.
Mike Trkay, chief customer officer and CIO of Fair Isaac -- which goes by the name FICO after its well-known credit score -- said its 3,500 employees are creating dozens of new AI agents daily.
"Every day there's quite literally new agents that are being created, and almost at every tier of the hierarchical structure," Trkay said.
Those agents are both "individual-level" bots for tasks like managing email or writing a brief, to larger scale agents that can manage data sets for multiple projects, he said.
The company is now instituting governance practices to prevent too many AI agents from providing conflicting results for the same problem. "We recognize the risk that it presents," Trkay said.
Employees at DaVita have created over 10,000 AI agents, said CIO Madhu Narasimhan. To help prevent cybersecurity risks, Narasimhan said the kidney-care company doesn't allow "consumer-grade" AI agent tools into its corporate environment. "Because we care for our patients, we have to scale with safety," she said.
DaVita also has developed an internal platform that allows the company to pull back on inference and manage token costs when needed, but also lets it increase token spend on the highest-performing AI agents.
Rounding up the bots
Some corporate tech chiefs are similarly taking a proactive approach to managing their AI agents, hoping to prevent agent sprawl from becoming a more widespread problem.
At Lyft, Jason Vogrinec, the ride-hailing company's executive vice president of AI transformation, said the company has rolled out Claude to its employees, and figured out an IT-approved way of sharing "skills, " the set of instructions that help Claude agents handle specific tasks.
Lyft also is working on creating a centralized platform with IT controls for all of its agents. Having too many agents creates a challenge for "a publicly traded organization with lots of regulatory obligations," Vogrinec said.
Anthropic said it has rolled out features that help IT admins in areas like role-based access, spend controls, usage analytics, audit logging and curated plug-in libraries.
Manu Narayan, GitLab's CIO, said the software-development tool maker's existing governance and guardrails are "holding the line" in preventing too much AI agent proliferation.
But there are also benefits of having too many AI agents -- at least for now.
"There will be agent sprawl in the short-term," Narayan said, "and we're actually OK with that because of the opportunity that AI presents."
Write to Belle Lin at belle.lin@wsj.com
(END) Dow Jones Newswires
May 15, 2026 07:00 ET (11:00 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments