AI agents are learning to work together — and companies must keep up

AI Agents Are Learning to Collaborate. Companies Need to Work With Them

With multiagent systems gaining traction, businesses should prepare to integrate coordinated AI agents across operations

Artificial intelligence is evolving — and it’s becoming collaborative. As developers push AI from individual agents toward multiagent systems that can operate autonomously and interact with one another, businesses must act now to prepare for this new AI frontier.

The rise of multiagent systems means enterprises will soon move from deploying isolated AI tools to orchestrating intelligent teams of agents capable of managing complex, interconnected workflows in areas like customer service, logistics, finance, marketing, and strategy.

“These are not isolated functions,” said Kathy Kay, CIO at Principal Financial Group. “They are systems of tasks that, when connected through intelligent agents, can drive faster insights and better outcomes across the enterprise.”

The next evolution: Agent-to-agent collaboration

While many organizations are still deploying basic AI tools, tech leaders like Accenture, Salesforce, and Google are building frameworks that allow agents to interact, reason, and negotiate with each other in real time.

Accenture has already created over 50 multiagent systems in use today across industries like automotive, media, and consumer goods. One such system includes 15 specialized agents, with three coordinating “super agents” managing a marketing campaign end-to-end — from research to content strategy.

“Only 10–15% of clients are using multiagent systems now,” said Lan Guan, Chief AI Officer at Accenture, “but we expect that to exceed 30% within 18–24 months.”

To support this growth, Accenture has launched Trusted Agent Huddle, a framework for agent-to-agent interoperability in collaboration with tech giants like Amazon, Google, Microsoft, Nvidia, Salesforce, and SAP.

Salesforce and Google develop A2A protocol

At the April Google Next conference, Salesforce and Google announced they are co-developing the A2A protocol (Agent-to-Agent) to allow intelligent agents within Salesforce’s Agentforce ecosystem to communicate and coordinate with both internal and external agents.

“It focuses on authentication, identification, and message passing,” said Gary Lerhaupt, VP of product architecture for Agentforce.

This protocol could standardize multiagent integration across platforms, potentially becoming the backbone of enterprise-wide intelligent automation.

From workflows to autonomy: What multiagents really do

Startups like Keyway, a commercial real estate tech firm, are already deploying semi-structured agent systems. Its platform uses coordinated agents to answer questions like how to price rental properties or what amenities to offer. But these agents still depend on predefined workflows and human oversight.

A true multiagent system, says Keyway CEO Matias Recchia, would feature agents that dynamically adapt to new information, collaborate without explicit prompts, and operate without human-defined sequences.

That’s the direction enterprise AI is heading — from tools that require oversight to systems that function as autonomous, strategic collaborators.

How companies can prepare for multiagent AI

The transition to multiagent systems won’t be instant. Experts recommend businesses begin now by:

• Deploying individual agents for standalone tasks.

• Building robust data pipelines to support real-time decision-making.

• Designing governance models that accommodate autonomous interactions.

• Updating workflows to integrate both human and AI collaboration.

At Principal Financial Group, AI agents already support domains such as software engineering, claims summarization, and post-call analytics. Kay says her team is now developing the infrastructure to enable agent-to-agent interactions for complex applications in asset management, retirement services, and contact centers.

From silos to systems: The enterprise AI shift

Multiagent systems promise to elevate AI from isolated point solutions to fully integrated digital ecosystems. In asset management, agents might analyze unstructured market data, generate investment narratives, and align strategies across portfolios. In customer service, they could coordinate claims handling, chat support, and escalation logic in real time.

According to Accenture, multiagent system adoption will continue accelerating through 2025, especially as frameworks like A2A mature and more companies look to AI for cross-functional decision-making.

As AI agents learn to collaborate and make autonomous decisions, organizations will need to move fast to align technology, workflows, and strategy. Those who adapt early will gain a decisive advantage in the emerging era of coordinated, intelligent enterprise systems.

Stay tuned to The Horizons Times for updates on enterprise AI, multiagent systems, and the future of business automation.