Big Four firm PwC launched what it styles as an agentic AI operating system to streamline AI workflows and orchestrate complex, multi-agent business processes at scale.
Dubbed "Agent OS," it acts as both the central nervous system and the switchboard for enterprise AI via a patent-pending orchestration system. The solution connects AI agents with each other, regardless of platform or framework, into modular adaptive workflows integrated with enterprise systems such as those from Anthropic, AWS, GitHub, Google Cloud, Microsoft Azure, OpenAI, Oracle, Salesforce, SAP, Workday and others. It also adapts workflows to multiple languages, supporting integration for global enterprise needs.
Matt Wood, PwC's commercial technology and innovation officer, said Agent OS is deployed within the client's environment, leveraging their own existing infrastructure, models and frameworks. Pre-built agent and workflow blueprints are configured and run entirely within the enterprise's own systems. This setup allows Agent OS to integrate directly with the client's infrastructure, models and data sources, so the data remains within their control, as all data processing takes place on their side: Agent OS accesses enterprise data, invokes models as configured by the client, and outputs results back into their systems.
Since the deployment is within the client's environment, governance and oversight fall under their internal frameworks, though PwC provides guidance and best practices to support responsible and secure implementation, said Wood. This also means usage-based costs would fall under the client's broader AI or cloud budget.
Agent OS will come built with a large library of pre-built, fully customizable AI agents, but if users don't like any of those, they can create their own in-house using third party software development kits already integrated with PwC's solution or fine-tuned on the user's own proprietary enterprise data.
The software uses a drag-and-drop interface as well as natural language transitions and data flow visualizations to ease workflow creation for both technical and non-technical users.
Prior to today's rollout, PwC itself has been using Agent OS across its tax, assurance and advisory services for clients, having deployed 250 different AI agents across the firm for specific tasks.
For example, Wood said PwC uses AI-driven Code Intelligence agents to automate code documentation, test generation and modernization, which helps reduce technical debt and optimize software development. Document Extraction and Processing agents extract relevant information from a document using vector search and user-defined search terms, and then determine if any formatting or post-processing steps are necessary to adjust the output and then formats the response into the correct format. Data Bridge agents extract structured and unstructured data from Excel files or PDFs, then map it to a desired target format, parse and validate the extracted values to then interact with the documents and full datasets to gather insights, ultimately compiling everything into a final CSV file. Hundreds of other agents are in production, streamlining PwC's ERP and other system implementations for clients.
Wood said PwC has been building and deploying AI agents at scale within PwC for over a year, which has allowed it to formalize agent development, deployment, orchestration, monitoring and governance standards into the product announced today.
During the testing period, PwC learned many key insights, one of which was discovering that having multiple agents operate within a single conversation stream could result in inconsistent or unpredictable outputs. In response, PwC introduced deterministic data routing between agents and their respective skills to improve performance and reliability. Additionally, said Wood, developers found that overly negative reinforcement could confuse agents and misdirect behavior, so they began incorporating structured prompts and positive reinforcement techniques to help guide agents more effectively within defined workflows.
This is the latest development in PwC's push to accelerate its technological development, having invested about
