ServiceNow AI Workflow Factory Pairs Agents With MCP and A2A

What ServiceNow announced

ServiceNow has introduced two linked offerings, AI Workflow Factory and Autonomous Engineer. The company presents them as a way for enterprises to stop treating AI as a series of isolated pilots and start treating workflow improvement as a continuous, governed cycle. 5 The announcement came on a Tuesday. Coverage across several outlets framed it as a shift from "AI projects" toward ongoing reinvention. 5

The Factory is less a single new product than a connective layer across existing ServiceNow tools. It strings together Process Mining, Autonomous Engineer, Build Agent, and App Engine. 1 In the company's description, Process Mining identifies processes worth fixing based on business metrics. The development tools then build and test changes, and App Engine runs the resulting workflows at scale. 1 The pitch is a closed loop: discover, build, deploy, measure, repeat.

Autonomous Engineer and the role of humans

Autonomous Engineer is the most aggressive piece. ServiceNow says it supports "unattended coding" for planning, building, and testing implementation work, while developers keep control over critical decisions. 1 In practice, the agent drafts an implementation plan. After a human approves it, the agent carries out the development and testing. 1

This fits a broader trajectory in ServiceNow's platform releases. In the Brazil release, Autonomous Engineer moves further away from preview status and toward being a structural part of how the platform handles development. 3 Agents can run multi-step development workflows, including writing code, running tests, and iterating on failures, without a person kicking off each step. 3 Build Agent, its human-assisting counterpart, is now the default rather than a preview. It also updates itself automatically on instances running Brazil Patch 0, Australia Patch 5, or Zurich Patch 12 and later. 3 Automatic upgrades reduce admin overhead. They also mean the tooling that writes enterprise workflows changes underneath customers on the vendor's schedule.

The interoperability plumbing

A loop that only touches ServiceNow's own data would have limited reach. That makes the company's Action Fabric framework relevant context. Action Fabric has three components. 2

• MCP Server Console exposes ServiceNow capabilities to outside AI clients and handles publishing, governing, and observing those tools. 2
• MCP Client lets ServiceNow agents call out to external MCP servers. 2
• A2A lets ServiceNow agents work with external agents as peers on shared tasks. 2

MCP, the Model Context Protocol, is an open standard for how agents discover and invoke tools in other systems. 2 ServiceNow's documentation on enabling MCP and A2A has been updated through Zurich Patch 4. The company pitches it as a way to give its agents best-of-breed external tools and to support cross-organizational agent cooperation. 4

ServiceNow's materials do not explicitly describe AI Workflow Factory as running on Action Fabric. My reading, though, is that the two are designed to fit together. A factory that discovers and rebuilds processes across an enterprise needs agents that can reach outside the platform. MCP and A2A are how ServiceNow intends to provide that reach.

One more change shapes the picture. Starting July 9, 2026, ServiceNow will begin rolling out third-party model providers as the default for out-of-the-box Otto skills and AI agents, tied to individual store app release schedules. 4 Taken together, the platform is becoming more open on several fronts at once: external tools, external agents, and external models.

Governance is the real question

Openness and autonomy widen the attack surface and the blast radius. CIO.com's coverage highlighted this. It quoted an analyst who argued that CIOs need the ability to revoke agent authority and to recover from a failure of the control plane itself. 1 The same coverage noted that an enterprise should be able to replace pieces of that control layer, a pointed concern for a vendor positioning itself as the central control tower. 1

ServiceNow's answer so far is human approval gates on Autonomous Engineer 1 and governance and observability built into the MCP Server Console. 2 Those are reasonable starting points. They mostly address what agents do inside approved boundaries, however. They do less to address what happens when the governing system fails, or when a customer wants to leave.

The takeaway

AI Workflow Factory is a coherent bet. Process mining tells agents where to work, agentic development does the work, and open protocols extend that work across the stack. The technical ingredients are largely already shipping. The harder challenge is organizational: making sure every loop has a reliable off switch.

Enterprises evaluating the Factory should treat agent revocation, audit trails across MCP and A2A connections, and control-plane portability as core requirements, not afterthoughts. How well ServiceNow answers those questions will determine whether "continuous reinvention" becomes durable infrastructure or a new kind of lock-in.