Workday Rising 2026: Claude, Sana, and the Headless Agent Platform
On October 13, 2026, at Workday Rising in Las Vegas, Workday and Anthropic announced a direct partnership connecting Claude’s reasoning to Sana, Workday’s agentic experience layer, via the Model Context Protocol. The official announcement came alongside three other major releases: a headless Trusted Agent Platform, Workday Work Intelligence, and a plan to bring Sana into Claude, Gemini Enterprise, Microsoft Copilot and Teams, and Slack. Together, the announcements lay out Workday’s position as the governance and data layer for enterprise AI across HR and finance.
More than 11,500 organizations, including more than 65 percent of the Fortune 500, run Workday for HR, finance, and IT.
The Anthropic Partnership: Claude Inside HR and Finance
The core of the Workday-Anthropic integration is bidirectional. Employees who work in Claude can now surface Sana agents, query Workday HR and finance data, and take action without leaving the conversation. At the same time, Sana can call Claude for complex reasoning at the right point inside a workflow, sharing only the context that step requires.
A few concrete examples from the announcement:
- A hiring manager in Claude can compare open roles against existing headcount and budget, then build an internal mobility plan using live Workday data, with the action recorded in Workday’s audit trail.
- A finance leader can review invoice exceptions against purchase orders and approve or route them to Sana, without opening a separate Workday tab.
- During a compensation cycle, Sana orchestrates the full process inside Workday, calling Claude only when it needs reasoning to draft a manager note, sharing only the context required for that specific step.
Every action in this model carries Workday’s existing permission layer. Claude does not receive raw Workday data. Sana mediates every request, applies the customer’s business rules and approvals, and surfaces only what the specific user is authorized to see.
The architecture relies on the Model Context Protocol. Workday exposes its Agent-Ready Tools as MCP-compatible endpoints. Any MCP-capable client, including Claude, can connect to those tools once, with no per-user setup and no new permission model to build.
Ankur Sinha, head of engineering for enterprise and verticals at Anthropic, described the practical implication: “Workday’s Agent-Ready Tools bring that model to HR and finance, helping customers give agents access to critical workflows while preserving the permissions and approvals their businesses already trust.”
The Sana in Claude plugin is available to early access customers today and is expected to be generally available in late 2026.
The Trusted Agent Platform Goes Headless
Simultaneous with the Anthropic announcement, Workday made its Trusted Agent Platform headless, meaning the full HR and finance capability stack is now accessible to any external AI tool or agent, not only to Workday’s own interface.
The platform is now generally available across five components:
| Component | What It Does | Status |
|---|---|---|
| Workday Data Cloud | Live SQL access to Workday data via Live Data Query; no stale copies, security enforced | Generally available |
| Agent-Ready Tools | MCP-compatible tools that expose HR and finance actions to any external agent | Generally available |
| Agent Harness | Runtime governance layer for all agents; provides context, policies, routing, model selection | Generally available |
| Developer Agent | Natural-language agent builder using Claude Code, Codex, Cursor, or Workday Developer Site | Generally available |
| Custom Agents | Deploy custom agents built with LangChain, Google ADK, or any preferred framework | Generally available |
Agent Harness deserves particular attention for enterprise buyers. Every agent deployed on Workday, whether built by Workday, a customer, or a partner, runs through Agent Harness. The harness gives each agent the context a task depends on: who the person is, their role, what they can see and do, which approvals apply. That context comes from the live system of record, not a policy document or data copy.
Agent Harness also governs action routing and model selection for cost and performance. Two capabilities are expected in early access by late 2026 and early 2027: Agent Context Service (pre-loading the context an agent needs before it starts reasoning) and Grounded Reasoning (checking requests against company policies before an agent acts).
For customers building on the platform, the model choice is open: Anthropic Claude, Google Gemini, Meta Llama, Mistral models, OpenAI GPT, and NVIDIA Nemotron open models are all supported. Support for LangChain frameworks and NVIDIA Nemotron is now generally available; support for Google ADK is accepting design partners.
This headless architecture matters for enterprise AI buyers who have been building their own orchestration layers on top of enterprise systems. Workday is effectively offering an alternative: instead of connecting agents directly to an enterprise system through custom APIs and replicating security logic in the agent layer, customers can connect once through Agent-Ready Tools and inherit Workday’s existing permissions and audit trail with no additional configuration. This is the same principle behind the multi-agent governance challenge enterprises face: who controls what the agents can actually do at runtime.
Work Intelligence: Mapping How AI Changes Jobs
Workday Work Intelligence is a new product now available in early access to Workday HCM customers. The core problem it addresses is that most organizations can see headcount and job titles but not the individual tasks that make up each role.
Work Intelligence breaks each job into tasks, maps those tasks to the skills they require using Workday Skills Cloud (75,000 canonical skills, used by more than 3,450 customers), and gives leaders a view of which tasks AI agents can take over, which they should augment, and which must remain with a person.
The output is a workforce planning tool that updates continuously as business needs and the capability of AI agents evolve. Three integrations extend the value:
- Decision Intelligence: Quantifies the time and cost advantages of developing skills internally versus hiring externally, once Work Intelligence has identified which capabilities the business will need next.
- Workday Learning powered by Sana: Turns skill gap analysis into personalized learning paths for individual employees, including upskilling recommendations tied to internal mobility opportunities.
- Ongoing measurement: Tracks how an employee’s time shifts as AI handles more of certain tasks, which skills are growing in demand, and whether employees are moving into new roles created by AI adoption.
This is the enterprise-side tool for what is increasingly called workforce redesign: the deliberate process of deciding, at the task level, which work AI should own, which it should support, and which belongs exclusively to humans. The framing is less “will AI replace jobs” and more “here is specifically which parts of this job are changing, and what do we do about it.”
Sana Everywhere: Claude, Copilot, Gemini, Slack
The third announcement from Workday Rising 2026 is Sana’s expansion into the AI tools employees already use. As of today, Sana self-service capabilities are generally available inside Google Gemini Enterprise, Microsoft Copilot, Microsoft Teams, and Slack. Full Sana agentic capabilities will arrive on those platforms in late 2026.
Sana in Claude Cowork is available today in early access. Sana in Microsoft Copilot Cowork is expected in early access in late 2026.
The technical distinction between self-service and full Sana matters. Self-service lets employees ask questions and get answers from Workday data inside these tools. Full Sana moves beyond answers: an employee describes an outcome, Sana coordinates the right Workday agents and tools behind the scenes, and the result is recorded in Workday. For example, after a business trip, Sana can flag an unfiled expense inside Claude, apply company policy, prepare the expense report for review, and let the employee confirm it from the conversation.
Customers do not need to configure each tool separately. A single connection to Sana brings all Workday agents, including custom agents built by customers and partners, into every connected platform.
Why This Matters for Enterprise AI Buyers
The Workday Rising announcements collectively represent something enterprise buyers have been asking for since agentic AI started reaching production: a governed execution layer for AI agents that does not require rebuilding trust and permission logic from scratch for every new tool.
Three aspects stand out for organizations with complex AI deployment roadmaps:
The MCP bet: Workday is putting the Model Context Protocol at the center of its enterprise integration story. Agent-Ready Tools are MCP endpoints. Sana uses MCP to exchange tasks with Claude. This is a deliberate technical alignment with an open standard that is gaining adoption across the AI industry, which means Workday’s integrations should work with any future MCP-compatible client.
Model agnosticism as a governance choice: By supporting Claude, Gemini, Llama, Mistral, GPT, and Nemotron on the same Agent Harness, Workday separates the question “which model should run this task” from the question “what is this agent allowed to do.” Governance is centralized; model selection is an operational decision made per task.
Anthropic as the first featured integration: Despite positioning the platform as model-agnostic, Workday chose Anthropic as the launch partner for the deepest integration. The Sana in Claude plugin and the Anthropic Blueprint for Hypergrowth Finance session at Workday Rising signal where the initial production experience will be most refined.
For enterprise teams evaluating how to connect AI assistants to the systems where real business decisions are made, the Workday Rising announcements from October 13 provide a detailed reference architecture. If your organization runs Workday for HR or finance, the Trusted Agent Platform is a concrete path to giving AI agents governed access to those systems without building a separate integration layer. Teams looking at the broader question of agentic AI deployment infrastructure, particularly the parallel evolution of managed agent platforms, can also see what OpenAI’s Managed Agents API approach offers by comparison.
To explore how an AI-native GTM or operations design could leverage governed enterprise agent access in your organization, book a call with the Enera team.