Companies Face A New AI Challenge As Agent Sprawl Takes Hold

Businesses have spent the past two years racing to adopt artificial intelligence. Now, some of the most enthusiastic corporate users are confronting a new problem: they may have created too many AI agents.

As AI tools become easier to build, customise and deploy, companies are beginning to see the rise of what is being described as AI agent sprawl. The term refers to the uncontrolled spread of artificial intelligence agents across an organisation, often created by different teams, for similar purposes, with limited central oversight.

The issue is emerging as platforms such as Anthropic’s Claude Cowork and open source orchestration tools like OpenClaw make it simpler for employees to create and coordinate AI agents. These tools can be powerful, allowing workers to automate tasks, speed up workflows and experiment with new ways of using artificial intelligence inside a business.

But the same accessibility that makes AI agents appealing is also creating a management headache.

Companies including Lyft, DaVita and GitLab are among those navigating the challenge of agent proliferation, according to reporting from The Wall Street Journal. Their concern is not that employees are using AI too much, but that AI adoption is becoming difficult to track, govern and secure.

The problem begins when nontechnical employees can create agents without needing extensive coding skills or direct involvement from IT departments. In theory, that democratises innovation. In practice, it can lead to dozens or even hundreds of agents operating across different business units, sometimes duplicating the same functions, accessing overlapping data or consuming computing resources without a clear owner.

For corporate technology teams, this creates several risks at once.

The first is visibility. If an organisation does not know how many agents are running, what they are doing or what systems they can access, it becomes much harder to manage risk. An AI agent designed for a harmless workflow could still interact with sensitive information, customer data or internal systems if permissions are not carefully controlled.

The second is security. Every new agent can become another point of exposure. Poorly configured agents, forgotten experiments or duplicate tools may create unnecessary pathways into corporate systems. In industries handling health, transport, finance or sensitive customer information, that risk becomes even more significant.

The third is cost. AI agents often rely on computing power, model access and cloud infrastructure. If multiple teams create agents to perform similar tasks, companies may end up paying for duplicated work at scale. What begins as a productivity experiment can quietly become a growing operational expense.

The fourth is accountability. When an AI agent produces an error, makes a recommendation or automates part of a workflow, businesses need to know who is responsible for monitoring it. Without clear ownership, agents can drift from useful tools into unmanaged systems that continue operating long after their original purpose has faded.

Yet companies are also cautious about overcorrecting. Heavy handed restrictions could discourage employees from experimenting with AI, slowing the very innovation businesses are trying to encourage. The challenge is to create enough structure to reduce risk without shutting down productive use.

That balance is likely to define the next phase of enterprise AI adoption.

Rather than simply asking how many employees are using AI, businesses may need to ask more precise questions. Which agents exist? Who built them? What data can they access? Are multiple agents doing the same job? How are they monitored? What happens when they fail?

For many companies, the answer may be a more formal governance model for AI agents. That could include internal registries, approval processes, access controls, cost monitoring, security reviews and regular audits to identify duplicate or outdated agents. Businesses may also need clearer rules around when employees can create their own agents and when IT or security teams must be involved.

The rise of agent sprawl shows how quickly AI has moved from novelty to infrastructure. What was once a small productivity tool is now becoming part of how companies operate, communicate and make decisions. That makes governance no longer optional.

AI agents promise faster work, smarter automation and new forms of productivity. But without oversight, they can also become messy, expensive and risky. For companies embracing the technology, the goal is no longer just to adopt AI quickly. It is to adopt it responsibly, sustainably and with enough control to know exactly what is happening inside the machine.

Photo Credit: DepositPhotos.com