October 20, 2025 Anthropic Claude Haiku 4.5 now available in PwC’s agent OS via Amazon Bedrock
Anthropic Claude Haiku 4.5 delivers near–frontier-level performance at a fraction of the cost and latency, enabling faster, cheaper, and more scalable agentic workflows
Support for Anthropic’s Claude Haiku 4.5 is now available in PwC’s agent OS via Amazon Bedrock, giving clients a fast and cost-effective model option for orchestrated AI workflows. Haiku 4.5 delivers near-frontier coding and agent performance at substantially lower cost and higher speed—a fit for scaled deployments, free-tier experiences and sub-agent patterns where latency and budget matter.
What’s new
Anthropic’s Haiku 4.5 provides similar levels of coding performance to Claude Sonnet 4 at roughly one-third the cost and more than twice the speed. This strong coding and agent performance, including computer-use tasks, measured at 73.3% SWE-bench, verified per Anthropic’s methodology.
What it means in agent OS
With Haiku 4.5 available through Bedrock, agent OS can route and coordinate more sub-agents per workflow at a given budget, helping teams decompose complex jobs and run parallel task execution (e.g., research threads, code refactors, customer-support actions) while maintaining governance in an enterprise environment. Anthropic highlights a recommended pattern where a planning model coordinates multiple Haiku 4.5 workers for parallel subtasks—an approach agent OS is designed to operationalize across tools and systems.
Model highlights relevant to enterprise orchestration
- Speed + cost profile suited to latency-sensitive and scaled agent use cases.
- Hybrid reasoning with near-instant responses or extended thinking; available via Amazon Bedrock for enterprise integration.
- Safety reviewed by Anthropic with ASL-2 release context per their public write-up.
Use cases to consider in agent OS
- Coding sub-agents: test generation, refactoring, CI prompts
- Research sub-agents: parallel document synthesis, source extraction
- Latency-sensitive experiences: assistants, support flows
- Browser / computer-use tasks coordinated within larger processes
Claude Haiku 4.5 is available now via the Claude API and major clouds, including Amazon Bedrock. Agent OS support through Bedrock is live for eligible deployments.
Sept. 23, 2025 Enterprise-grade security enhancements
As agent OS continues to scale across enterprise environments, we’ve introduced new security features to help teams manage risk, maintain governance, and meet internal compliance standards—without slowing down development.
Explore the newest capabilities:
- Consistent role-based access control (RBAC): Access policies for agents, tools, and workflows now apply uniformly—no matter the cloud provider or framework. Define access once, and agent OS applies it across environments.
- Group administrator role: Teams can now manage their own users and permissions directly, enabling faster collaboration while keeping system-wide controls intact.
- Secrets management: Agent OS supports integration with cloud-based Key Vaults, allowing tools and agents to use organization-managed credentials. This confirms credentials remain within governed cloud infrastructure, a necessary requirement for regulated and zero-trust environments. Key usage is scoped by role and fully auditable.
- Isolated tool execution: Custom tools now run in sandboxed environments with no public network exposure, strict resource limits, and built-in security scanning. This design can reduces the risk of cross-process interference and limits impacts from potential vulnerabilities — helping enterprises run AI-powered workflows with greater confidence in security, stability, and governance integrity.
Whether you're preparing for audits, securing sensitive use cases, or aligning with internal security standards, these updates help organizations leverage agent OS to move from experimentation to enterprise-scale adoption—with the guardrails to support even the most risk-sensitive deployments.
August 8, 2025 PwC expands agent OS with GPT-5 support for next-generation AI orchestration
PwC is expanding agent OS, its enterprise AI orchestration platform, to support OpenAI’s newly released GPT-5 model. The integration enables enterprises to deploy autonomous agents with enhanced planning, reasoning and tool use capabilities across business workflows and cloud environments.
The first AI orchestration platform to support GPT-5 models, PwC’s agent OS is an enterprise AI command center that unifies and orchestrates intelligent agents into modular, adaptive workflows—enabling organizations to build, integrate, and scale AI-driven processes up to ten times faster than traditional methods. Clients value agent OS for its seamless integration across platforms, built-in governance and risk management, and its proven ability to help deliver measurable efficiency gains and ROI—such as reducing call-center times by 25% and shortening compliance review workloads by as much as 94%.
In addition, GPT-5’s advanced reasoning allows PwC to help reduce the number of agents required for code-generation use cases — streamlining orchestration, improving latency and simplifying agent management. Early results show that fewer, more capable GPT-5–powered agents can replace validation and judging agents without sacrificing quality.
Moreover, GPT-5 delivers substantially improved quality when generating long, complex documents — such as functional specifications, technical specifications, process design documents — and excels at large-scale reasoning tasks. In production, millions of agent interactions powered by PwC’s agent OS will benefit from greater accuracy with the improved reasoning model, driving faster time-to-insight and higher productivity.
With GPT-5, agent OS continues to provide organizations with a flexible, governed framework for integrating large language models into end-to-end business processes. New capabilities from OpenAI—such as GPT-5—will continue to be delivered through PwC’s agent OS in alignment with OpenAI’s general availability schedule, further accelerating the creation of agentic blueprints. This momentum will help drive rapid transformation across finance and HR functions, as well as industry-specific blueprints for utilities and energy, financial services and beyond.
This integration builds on PwC’s participation as an OpenAI Services Partner in early access testing of prior models. These testing efforts help inform our orchestration design around emerging capabilities and allow us to share structured feedback with OpenAI to help drive continuous improvement. PwC will continue to test GPT-5.0 and ChatGPT agents under secure sandbox conditions — further supporting our goal of enabling responsible, scalable AI deployments across industries.
July 15, 2025 PwC expands agent OS to AWS, bringing Bedrock and AWS-native tools to enterprise AI workflows
PwC has announced a major expansion of its enterprise AI orchestration platform, agent OS, with support for Amazon Web Services (AWS) environments and native integration of Amazon Bedrock. This enhancement enables organizations to run agent OS within AWS and orchestrate AI workflows using Bedrock-hosted models alongside AWS-native tools —without the need to refactor for another cloud provider.
What’s new
The AWS-compatible release introduces a set of integrated tools and execution support designed to enhance enterprise AI workflows in native AWS environments:
- Amazon Textract: Uses optical character recognition (OCR) to extract structured data (forms, tables) from PDFs and scanned documents, supporting pre-processing steps in intelligent workflows.
- Amazon S3: Allows scalable, secure file storage and retrieval within Amazon S3 buckets, enabling secure dynamic data access.
- Amazon OpenSearch Service: Power high-performance search across structured and unstructured data, supporting full-text queries, filtering and aggregations.
- Amazon Bedrock agent framework: Connects agent OS to Amazon Bedrock for invoking foundation models (e.g., Amazon Nova), enabling document drafting, translation, summarization and more.
- MCP server support: Makes AWS-connected tools callable from within custom workflows using the MCP protocol — enabling seamless tool calling for agents.
June 11, 2025 PwC expands agent OS to Oracle Cloud, enabling cross-platform AI orchestration for enterprise workflows
PwC today announced the latest expansion of its enterprise AI orchestration platform, agent OS, with full support for Oracle Cloud Infrastructure (OCI) and Oracle Fusion applications.
Our OCI agent OS support introduces several new capabilities and extensions within the agent OS framework:
- Integration with Oracle's generative AI services, allowing agent OS to access Oracle-hosted large language models and embedding services. This opens the door to more intelligent agent interactions, including summarization, extraction and conversational workflows — powered by Oracle's AI platform.
- Availability of new Oracle-native tools within agent OS, allowing agents to use tools like the following to accomplish objectives and tasks:
- The Oracle Cloud Storage tool, which enables AI agents to store, manage and retrieve data directly from OCI — supporting secure data movement across enterprise workflows.
- The Oracle Vector DB tool, which integrates Oracle's high-performance vector search capabilities into agent OS workflows — enabling advanced document retrieval and context-aware agent responses.
- Support for MCP server-based tool execution, making Oracle-connected tools callable, following the MCP protocol, from within custom workflows. This enables organizations to build once and reuse Oracle-integrated capabilities broadly across departments and use cases.
- Deployment flexibility, with agent OS now able to run directly in OCI environments and interoperate with Oracle Fusion applications, giving teams full control over how and where their AI agents operate.
May 16, 2025 PwC Unveils Agent Powered Performance: A Built-In Business Engine That Finds What’s Broken and Helps Fix It Fast
Today, PwC announced the launch of Agent Powered Performance, an AI-enabled business performance engine that helps companies quickly find hidden opportunities, act faster, and drive measurable results. Built for leaders across operations, finance, technology, growth, and strategy, it enables organizations to reduce inefficiencies, boost productivity, and make smarter decisions—without overhauling their systems. Unlike traditional tools or one-off initiatives, Agent Powered Performance combines continuous AI-driven insight with embedded execution, helping companies improve every day—not just during major transformations.
Reimagining Business Performance for Faster Value Realization
Today, most businesses find problems only after they’ve already hurt performance. A finance team might not notice billing errors until they show up in quarterly reports. A supply chain team might not catch rising shipping costs until customer complaints start rolling in. By then, it’s too late—and the fix often takes weeks. Agent Powered Performance changes that by helping businesses see problems earlier, fix them faster, and keep improving every day.
Agent Powered Performance is backed by PwC’s “Sense, Think, Act” approach supported by PwC's patent-pending AI agent orchestration platform, agent OS—a continuous cycle that helps businesses detect issues earlier, make smarter decisions, and execute faster. AI agents embedded in enterprise systems sense where value is leaking, think through the most effective performance strategies using predictive models and industry benchmarks, and act directly in tools like ERP or CRM to make improvements stick. Because it uses pre-built models tailored to each sector, companies don’t have to start from scratch. The system connects directly into ERP environments, continuously monitors key metrics, and acts inside the tools teams already use. For example, a supply chain agent might detect rising shipping costs and automatically reroute deliveries to reduce spend. Finance agents can spot and correct billing errors before they reach the customer. Clients typically see measurable efficiency gains in the first quarter, with continued improvements over time as the system learns and adapts.
May 16, 2025 PwC adds support for Model Context Protocol in agent OS
Today, we are announcing support for the Model Context Protocol (MCP) in PwC’s agent OS—unlocking a secure, scalable way for AI agents to access the tools and data they need to act. This integration bridges two critical layers of enterprise AI, intelligent agents and the systems they rely on to deliver real outcomes.
Connecting AI agents to action
As organizations scale their AI efforts, agent-based systems have become the preferred model for embedding intelligence into business workflows. These agents can reason, act and, when orchestrated effectively, collaborate to solve complex tasks.
But to move from experimentation to execution, agents need more than intelligence. They need structure. That means three things: governance, orchestration, and access. Agent OS already provides the first two. With the addition of MCP, it now offers the third—secure, standardized access to enterprise tools.
New capabilities through MCP
This integration unlocks a set of core capabilities that make agent systems more practical to build, easier to manage, and safer to scale. First, it enables reusable tool access across the entire environment. Once an agent system is registered as an MCP server, any authorized agent can make use of it. This eliminates redundant integration work and the overhead of writing custom logic for each new use case.
Second, it accelerates the development process. By standardizing how agents invoke tools and handle responses, MCP simplifies the interface between agents and enterprise systems. This consistency reduces development time, lowers testing complexity, and cuts deployment risk. Teams can spend less time on infrastructure and more time on business logic.
Third, governance is built in from the start. An interaction between an agent and an MCP server is authenticated, authorized, and logged. Access policies are enforced at the protocol level, which means that compliance and control are native to the system—not layered on after the fact.
