AI agents orchestration with Azure AI Foundry: PwC

Summary

AI agents are moving beyond simple chatbots, but struggle across fragmented apps, data, and partners.
Orchestration gaps create security, governance, and visibility risks.
Microsoft Azure AI Foundry plus PwC’s agent OS create a cross-cloud operating layer with unified controls, monitoring, and workflows to help scale agents and drive measurable gains.

Artificial intelligence is dramatically reshaping your enterprise. But somewhere between marketing hype and success stories lies an often-overlooked fact. As organizations evolve beyond chatbots and basic tools to full-fledged AI agents, they often encounter difficulty getting their AI to work across broad ecosystems of applications, data, and partners.

AI agents can automate multi-step workflows and processes—producing both quantitative and qualitative improvements. They can squeeze out inefficiencies and trim costs. However, building and scaling these autonomous systems involves obstacles, including fragmented systems, inconsistent orchestration, security risks, observability gaps, and governance requirements.

What’s needed is an orchestration layer that spans ecosystems and coordinates complex business workflows. With this operating layer in place, an enterprise can construct an agentic framework with embedded controls for safety, compliance, and ongoing improvement. This framework can reduce cycle time, lower costs, and help improve customer outcomes. It can build confidence and help unlock measurable business value.

Azure AI Foundry coupled with PwC’s agent OS delivers a consistent, scalable, cross-cloud platform that’s ideal for developing and embedding AI across your enterprise. With it, teams can simplify integration, establish powerful security and observability controls across clouds, and scale agentic transformation across business functions.

How agents change the equation

While AI agents can deliver transformational results, they introduce a practical challenge: few organizations operate on a single technology stack. Even a Microsoft-first enterprise relies on industry-leading SaaS platforms and specialized business systems—ERP, CRM, contact center tools, multiple data platforms, and custom business applications.

Bridging these systems is key to getting more value out of AI and agentic transformation. However, with agents running in different environments and data spread across applications, connecting tasks and workflows can prove daunting. In many cases, agents stretch across customer experience (CX) teams and contact centers. They interact with several platforms and data sources, such as SAP, ServiceNow, or Salesforce. This leaves no one in charge of orchestrating these agents and data access across providers.

Enterprises also encounter an operational gap: they can monitor agents inside individual applications, but they can lack visibility and control over end-to-end agentic workflows. If end-to-end processes break, it can be difficult to see who took which actions, what changed, and what the downstream effect was.

Not surprisingly, as more agentic solutions appear across your organization, governance becomes more difficult. Operational safety, identity and security, auditing, and data boundaries can blur—and enforcing controls across systems can become more difficult. What’s needed is a clear governance plan. Organizations should stay focused on three imperative factors: security, observability, and AI safety.

What Microsoft and PwC deliver as an agent engine

Scalable agents require powerful orchestration. Together, agent OS and Microsoft Azure AI Foundry provide the operating layer that’s required for secure end-to-end agentic reinvention.

Azure AI Foundry provides tools to design, deploy, and scale agents with built-in governance and monitoring. Its Agent Service REST API pairs LLMs with tools to read data, call functions, and execute logic without teams managing infrastructure. Model Context Protocol (MCP) then helps these agents connect to the tools and data they need in a controlled and secure way.

Copilot Studio adds a strong set of tools for building low-code agents and publishing them into channels like Teams and Microsoft 365 Copilot. This makes it possible to push AI across the Microsoft ecosystem, and distribute agents into app stores, catalogues, and marketplaces.

PwC’s agent OS accelerator complements these Microsoft capabilities by helping organizations develop agents and stitch them together into end-to-end workflows. It provides consistent security, governance, and observability across LLMs, agent frameworks, tools, data stores, MCP servers, and A2A servers. Agent OS accelerator isn’t a replacement for Azure AI Foundry (or other agent platforms); it’s an enterprise orchestration and governance layer that helps unlock utility and value from AI across a heterogeneous provider ecosystem.

How PwC agent OS serves as an operating layer for multi-platform agents

Together, Azure AI Foundry and agent OS deliver industry-leading results:

Complete orchestration and a single pane of glass. As functions such as CX, finance, and HR adopt AI agents, using multiple cloud and third-party solutions, they gain a unified view of activity across workflows—including agent actions, human interventions, outcomes, and downstream effects. This visibility makes it easier to understand events across systems and departments—and identify problem points as well as industry-leading practices.
Close integration gaps and accelerate time to value. Azure AI Foundry and agent OS can deliver immediate results, but the true value accrues as organizations build out agents and refine workflows. These enterprises evolve beyond basic shortcuts; they can achieve a stronger agent-based framework that ties together groups of once-discrete tasks. With this level of AI automation in place, transformational change follows.
Improve standardization and reuse. Azure AI Foundry helps teams build agents quickly, but agents can deliver greater value when they operate safely and smoothly across business systems. When agents retrieve a customer’s record, open a case, update an order, or trigger a workflow, it’s imperative that they produce consistent outcomes and controls. With agent OS orchestrating tasks, organizations establish a common format for managing, governing, and reusing components across their enterprise—as well as across multiple technologies, applications, and cloud providers.

What results are possible with agent OS and Microsoft?

It’s easy to fall into the trap of building out AI for the sake of AI. The downside to this is organizations potentially rolling out new features without fully understanding how they reshape business workflows—or whether they deliver measurable value. PwC and Microsoft focus on a business outcomes-first design. In PwC engagements, we’ve seen up to 90% efficiency gains for work order planning in power plants, and up to 40% efficiency gains in finance processing, such as procure-to-pay and order-to-cash.

These findings echo other PwC research. Our 2025 PwC Customer Experience Survey found that 94% of customers now rate seamless customer experience somewhere between a minimum expectation and a bonus, and 97% describe quality customer service as essential. PwC research also found that AI and data can reinvent customer experience and drive revenue transformation.

Establishing a path to success

If you’re looking to put AI agents to work and up your organization’s AI IQ, here are a few things to focus on:

Start with the entire business workflow or process, not just a single task. It’s wise to select two or three processes that involve manual handoffs, latency, or rework and redefine the full process with agents. From there, it’s possible to expand the initiative across other business units and, ultimately, the entire enterprise. Examples include areas like service resolution, finance close, procure-to-pay, and work order planning.
Build agents where the work exists. Identify high-value areas that can benefit from agents, and introduce metrics to gauge adoption and real-world impacts.
Design for a multi-platform reality. Assume that agents can interact with more than solely Microsoft systems. This can include ERP, CRM, and other industry platforms, sometimes hosted by other cloud providers.
Make governance and observability first-class. As the use of agents grows, treat identity, permissions, auditing, and end-to-end observability as table stakes—not something to deal with later.

Putting agents to work

As AI advances and organizations address more difficult tasks, AI agents are rapidly moving into mainstream enterprise use. These increasingly smart and context-aware systems help deliver real-world benefits by reducing the time it takes to complete business processes, improving productivity and decision-making, and lowering costs.

But success is far from guaranteed. It’s often difficult to convert complex business processes and intricate workflows into AI-powered automation. PwC’s agent OS—together with Microsoft Azure AI Foundry and Copilot Studio—offers a practical and affordable way to orchestrate agents, embed powerful controls, and, in the end, unlock the holistic transformational power of AI.