PWC’s recent move to integrate the Model Context Protocol (MCP) into its agent OS (an enterprise AI command centre) marks a meaningful evolution in enterprise artificial intelligence. By allowing intelligent agents to access tools and systems in a standardized and governed way, PWC is helping shift agent-based AI from concept to real-world execution. It’s a structural change that addresses one of the biggest enterprise AI challenges, scaling with control.
From Passive Agents to Active Operators
AI agents have long promised to vastly improve progress by performing tasks, making decisions, and collaborating with humans or other agents. However, until recently, much of that potential remained locked behind fragmented infrastructure and governance concerns.
PWC’s integration of MCP directly addresses this friction. The protocol allows agents to safely act within these enterprise environments by providing a consistent, reusable interface for agents to use tools, access data, and complete tasks across different systems.
By managing this access through agent OS, PWC embeds compliance, orchestration, and security into the architecture itself, rather than adding it simply as an afterthought. This is crucial in regulated industries, like finance and healthcare, where auditability and policy are non-negotiables.
Recently, we had an in-depth conversation with Scott Likens, the Global Chief AI Engineer at PWC, during which he shared his expertise on enterprise AI implementation in 2025.
Building on Broader Momentum
PWC’s move is part of a broader industry push towards more mature, agent-driven architectures. Across the enterprise AI landscape, leaders are laying the groundwork for scalable, multi-agent ecosystems.
- IBM has recently updated its Watsonx Orchestrate platform with new hybrid integration capabilities via webMethods. These updates allow enterprises to break down silos and connect disparate data sources, enabling more complex, cross-functional AI-driven automation.
- USAA is taking a cautious approach to their adoption of Watson x Orchestrate which they are using to automate sensitive data management tasks. Every agent interaction is monitored to ensure compliance with strict risk and compliance guidelines, a nod to the enterprise need for governance-first AI implementations.
- Microsoft, meanwhile, is incorporating AI agents from Anthropic’s Claude models into GitHub. This move allows developers to fix bugs and write code with the help of intelligent, task-specific agents. The addition of Anthropic’s models highlights the demand for multi-agent collaboration in development workflows, as well as the growing appetite for diverse AI frameworks in established platforms.
- Adobe has also entered the agent orchestration space with its Experience Platform Agent Orchestrator. This enables complex, multi-agent decision-making. Its architecture supports third-party agents and is tailored for dynamic, customer-facing environments like marketing and CX.
Why Governance-First Architecture Matters
What sets PWC’s approach apart is its emphasis on deeply embedded governance at the protocol and OS level. Key features include:
- Code-level security audits: All MCP servers undergo automated and manual code reviews.
- Credential management: Secrets are never written to disk but are securely injected at runtime.
- Role-based access control: Only authorised identities can invoke tool access, and all actions are logged in real time.
This level of rigor enables scale. In the absence of strong controls, AI agents acting across systems represent a massive security risk. PWC’s model helps enterprises sidestep this risk while gaining the operational benefits of agent-based AI.
Looking Ahead: From Pilot to Production
Integrating standardised protocols like MCP into enterprise architectures significantly changes enterprise AI landscapes. As agent-based systems mature, the emphasis is no longer on just what AI can reason, but on what it can safely do within governed environments.
Enterprises across industries are now tasked with turning experimental agent pilots into production-ready systems. That transformation depends on more than innovation alone; it requires robust frameworks for interoperability, orchestration, and of course, governance.
The future of scalable AI will not be built on individual breakthroughs but on architectures that allow many agents across many systems to collaborate securely and transparently.
