Enterprise Apps Unpacked
View All Series Articles
Podcast

Plan a multi-agent orchestration framework for scalable AI

In this podcast, 10Pearls partner Peter Hesse explains what can go wrong when expanding agentic AI across the enterprise, and why it helps to develop a governance framework.

David Essex
By
Published: 23 Mar 2026

Implementing agentic AI has become a priority for enterprise application vendors and owners, but real-world deployments have been limited in scope. In its November 2025 "State of AI" report, McKinsey found only 23% of organizations were scaling agentic AI somewhere in their organizations, most were only doing so in one or two business functions, and 39% were just experimenting.

Expanding agentic AI throughout an organization is, in large part, a problem of multi-agent orchestration, where AI agents communicate with each other and coordinate their activities across workflows, applications and even corporate firewalls.

Some software developers are recommending the adoption of multi-agent frameworks to manage the problem. Organizations can develop their own frameworks or get them from a variety of commercial and open source developers, including AWS and Microsoft.

In this episode of Enterprise Apps Unpacked, Peter Hesse, a partner at 10Pearls, a custom software development and consulting company, lays out the steps for developing a multi-agent orchestration framework. He explains why challenges with security, regulatory compliance and performance multiply when agents are deployed on a large scale.

Peter Hesse, 10Pearls partnerPeter Hesse

Rules for agents to follow

Hesse elaborated on a 10Pearls blog post that details the most common problems in deploying agentic AI at scale and the essential elements of an effective framework.

Security is a major concern because agents must talk to each other without compromising security. "Think about all of the things that have to do with security -- the triad of confidentiality, integrity, and availability of information," Hesse said. "You want to make sure your communication mechanisms are hardened for that."

He said 10Pearls recommends establishing a policy enforcement layer to ensure that business rules and compliance policies are applied equally across all agents in every workflow. "This saves you from having to write the same code into each agent, but instead you have a singular kind of framework or layer to pass everything through to make sure everything's following the rules."

Agentic AI has the potential to change how enterprise applications are developed, according to Hesse.

"It's going to become less focused on a screen-by-screen user journey approach and more around the creation of services and APIs that are meant to be called by agents, not by people," he said. "It means we have to change the way we think about doing these things and build things more modularly and that are driven more by events rather than mouse clicks and key entries."

It also means exposing data and workflows that were never intended to be exposed in the old style of developing applications.

Other topics discussed in the podcast include the following:

  • How a framework can support observability and traceability of agents.
  • Why agents' tendency to work in harmony with each other can make them less resilient in new, large-scale scenarios, and ways to counteract that.
  • Who should be responsible for developing a multi-agent framework.
  • Why being a "double bottom line" company is good for business.

David Essex is an industry editor who creates in-depth content on enterprise applications, emerging technology and market trends for several Informa TechTarget websites.

Dig Deeper on ERP administration and management