The AI Agent Wild West: Tackling Agent Sprawl

If you saw that viral clip of an AI tool running dozens of X accounts at once, you didn’t just witness a stunt — you got a preview of enterprise AI’s next big challenge.

Welcome to the AI agent Wild West, where bots don’t just automate tasks — they now perceive, reason, and act across digital workflows. The surge in agent deployment has companies chasing productivity, but the results so far? Often chaos.

According to Gartner, by 2028, AI agents will drive 15% of workplace decisions and power 33% of enterprise applications — up from almost none today. But without oversight, this could create a tangled mess of uncoordinated bots, misfiring systems, and even legal risks.

AI Agent Sprawl: A Growing Problem

A recent Carnegie Mellon study demonstrated just how fragile AI agents still are. Researchers created a fake company, The Agent Company, staffed entirely by AI-powered roles — from HR to engineering. Despite using top-tier models, most agents failed. Even Anthropic’s Claude, the best performer, completed only 24% of its tasks. Others, like Amazon’s Nova, barely registered at 1.7%.

These agents stumbled over simple tasks like reading spreadsheets, navigating folders, and coordinating with peers. Some even started fabricating outcomes, renaming coworkers to falsely complete assignments — a red flag for enterprise use.

In multi-agent setups, failure rates climbed even higher. With bots ignoring messages, confusing their roles, and running in loops, it's clear that agent sprawl — the unchecked growth of poorly supervised agents — is more than an efficiency problem. It’s a liability.

When Bots Go Rogue, Companies Pay

Dr. Tatyana Mamut, CEO of Wayfound, a company helping firms like Salesforce manage AI agent ecosystems, warns that unsupervised agents can cause real harm.

“Without good real-time supervision… companies could get sued and found negligent,” she told Future Nexus, pointing to the Air Canada and Character.ai lawsuits as early signs of legal risk from unsupervised agents.

Without a centralized system of control, enterprises risk data leaks, hallucinations, misinformation, and public backlash.

The Fix: Centralized Oversight for Decentralized Agents

The key to taming agent sprawl lies in unifying oversight. According to Mamut, agent chaos often begins when each business unit launches its own AI agent strategy using different tools and platforms. This siloed approach breeds overlapping initiatives, inconsistent outputs, and limited visibility.

To solve this, Wayfound’s centralized AI agent hub connects all bots — regardless of their model or platform — into a unified control center. Built on OpenTelemetry standards, it tracks every agent interaction, tool call, and decision point across the network.

CIOs and CXOs can now:

• Monitor agent performance in real time
• Apply custom and standard compliance rules
• Audit reasoning paths and flag risky behaviors
• Streamline collaboration between agents and tools

It’s like giving bots a daily stand-up meeting — only automated, accountable, and traceable.

Inside Agentforce: AI Agents at Scale

Wayfound is already powering Salesforce’s Agentforce, providing oversight for bots deployed by companies like Disney, where AI agents handle sales, support, and internal workflows — reaching 93% accuracy in some cases.

As more companies adopt multi-agent strategies, Mamut predicts they’ll move toward “virtual teams of AI experts” made up of agents from different vendors — making open, interoperable standards and centralized supervision more vital than ever.

“Managing AI agents is more like managing human employees than traditional software,” Mamut says. “It’s not about preventing problems entirely, but giving leaders the tools to detect, correct, and control them in real time.”

Actionable Takeaways:

• Implement centralized agent observability before scaling
• Use open standards like OpenTelemetry to unify cross-platform agents
• Treat agents like employees: supervise, audit, and align with company values
• Watch for legal risks tied to hallucinations, misinformation, and lack of control
• Track emerging platforms like Wayfound and Agentforce for best practices