Every Enterprise Is Building a New Workforce

Not another application. Not another chatbot. Not another productivity tool.

A workforce.

Across HR, Finance, IT, Operations, Customer Service, Marketing, Engineering, Legal, and Supply Chain, AI agents are beginning to perform work that previously required people. Some draft proposals. Some reconcile invoices. Some investigate incidents. Some coordinate supply chains. Some monitor compliance. Others collaborate with humans to execute increasingly sophisticated business processes.

This represents a structural shift in enterprise computing. For decades, organisations built operating models around human workers supported by software. Today, organisations are building hybrid workforces where humans and AI agents work together toward common business outcomes.

Microsoft calls these organisations Frontier Firms. Microsoft's 2025 Work Trend Index described their emergence. The 2026 report sharpened the challenge: employees are moving quickly, but organisational readiness is not keeping pace. IDC projects more than 1.3 billion active enterprise agents by 2028.

I believe this is one of the most significant shifts in enterprise technology since cloud computing.

The central question is no longer, "Should we deploy AI agents?" It is, "Can we see, govern, secure, and operate the agents already entering the enterprise?"

The First Enterprise AI Crisis Is Already Emerging

Most organisations don't have an AI problem. They have a visibility problem.

Over the past year, I've spoken with CIOs, CISOs, Chief Digital Officers, and AI leaders across industries. Almost every conversation eventually reaches the same question: 

Very few confidently answer yes.

Agents are being created inside Copilot Studio, Azure AI Foundry, partner platforms, open-source frameworks, departmental initiatives, hackathons, and citizen development programmes. Many solve real business problems. Many also operate outside formal governance.

This is what I describe as Agent Sprawl.

Unlike application sprawl or cloud sprawl, Agent Sprawl introduces a new class of enterprise risk — because these systems don't simply store information. They reason. They decide. They act. Often autonomously.

 

Why Traditional IT Controls Are No Longer Enough

Enterprise IT has historically governed four primary assets: people, applications, infrastructure, and data. AI agents fit neatly into none of these categories.

They possess identities. Access enterprise systems. Invoke tools. Collaborate with other agents. Execute workflows. Make decisions. Adapt based on changing context.

An AI agent is neither simply software nor simply a user. It is a new enterprise actor.

That changes everything: identity, governance, security, compliance, observability, lifecycle management, risk management, and business accountability all require rethinking.

Their effective risk profile is shaped not only by code, but by prompts, models, data, permissions, tools, memory, orchestration, and the tasks they are allowed to perform. An agent in week one may behave very differently by week six. That is the part that mobile management — and every prior IT governance model — never had to handle.

 

Enterprise AI Operations: The Next Enterprise Discipline

Every major technology shift has eventually created a new operational discipline. 

 

Cloud computing  →  CloudOps

Continuous delivery  →  DevOps

Security  →  SecOps

Data  →  DataOps

Financial governance  →  FinOps

AI  →  Enterprise AI Operations

 

This discipline is about much more than deploying models or building agents. It encompasses the full set of capabilities required to manage an AI workforce responsibly:

▸  Discovering every AI agent across the enterprise

▸  Assigning trusted identities and clear ownership

▸  Governing permissions, policies, and data boundaries

▸  Monitoring behaviour, outcomes, and anomalies

▸  Managing the full agent lifecycle from design to retirement

▸  Measuring business value - and acting on what the data shows

▸  Continuously optimising performance, cost, and risk

 

The organisations that master Enterprise AI Operations will scale AI faster - with less risk and greater business value. This is what WinWire AgenticOps is built to deliver.

Microsoft Agent 365: The Enterprise Control Plane

Rather than introducing an entirely new governance model, Agent 365 extends Microsoft's existing identity, security, compliance, and management capabilities to AI agents. Its role is not simply to manage AI — it is to provide an enterprise-wide control plane for the AI workforce.

Just as Microsoft Intune became the management layer for endpoints, Agent 365 is becoming the management layer for enterprise AI agents. As of May 1, 2026, it is generally available — and it reaches beyond Microsoft platforms to govern agents built on AWS Bedrock, Google Gemini, LangChain, and OpenAI. If an agent speaks Model Context Protocol, Agent 365 can see it, govern it, and secure it.

Core capabilities include centralised agent discovery and registry, Entra Agent ID, policy-based governance, Microsoft Defender integration, Microsoft Purview integration, cross-platform visibility, Model Context Protocol support, runtime protection, and agent observability.

 

Six Foundations of Enterprise AI Operations

These are not features. They are the foundational capabilities required to govern AI at scale - now shipping in Microsoft Agent 365.

1. Discover and register every agent — including the ones you didn't know about. Governance begins with visibility. A centralised registry should reveal which agents exist, where they were built, who owns them, what purpose they serve, which systems they connect to, and whether they are approved, restricted, dormant, or ready to retire. Discovery must include both sanctioned agents and shadow AI.

The Agent 365 Agent Registry now surfaces unmanaged local agents discovered automatically by Microsoft Defender, Entra, and Intune working together — covering more than 35 known agent types, including coding agents, AI desktop applications, and local and remote MCP servers. The registry doesn't just catalogue what IT registered. It finds what IT didn't know about.

2. Establish identity as the control boundary. Every agent should have a unique, governed identity — not borrowed credentials, shared secrets, or ambiguous delegation. Identity enables least-privilege access, Conditional Access, lifecycle management, ownership, authentication, and auditable accountability. In an agentic enterprise, identity becomes the practical perimeter.

Entra ID Governance now extends to ecosystem partner agents integrated through the Agent 365 SDK — enabling enterprises to apply access packages, lifecycle workflows, and least-privilege principles consistently across both internally-developed and third-party agents. A compromised agent gets contained in minutes rather than weeks, with a full audit trail for the day a regulator asks.

3. Govern access, data, and action. Knowing an agent exists is not enough. Enterprises need policy that defines what the agent may see, which tools it may invoke, which actions it may take autonomously, where human approval is required, and how sensitive information is protected. Governance must travel with the agent across systems and interactions.

Purview sensitivity labels, DLP policies, and audit records now extend to agent interactions. Microsoft Defender provides asset context mapping for each agent — the devices it runs on, the MCP servers configured for it, the identities associated with it, and the cloud resources those identities can reach — giving security teams the context to assess potential blast radius before an incident, not after.

4. Observe behaviour and operational health. Traditional telemetry tells you whether a service is available. Agentic observability must go further: what the agent attempted, why it acted, which data and tools it used, whether its behaviour drifted, whether a policy intervened, and whether the outcome was correct. This is essential for security response, reliability, auditability, and continuous improvement.

Security teams can now run proactive Advanced Hunting queries in Microsoft Defender to identify agents with risky configurations - such as MCP tools operating under maker credentials that allow privilege escalation - before they become incidents. Real-time visibility now means seeing the risk profile, not just the activity log.

5.  Manage the full lifecycle - including governance at the point of development. Agents should be onboarded, tested, approved, deployed, monitored, updated, reassigned, suspended, and retired through deliberate lifecycle controls. Clear ownership matters especially when an agent's creator changes roles, a business process changes, or the data and tools supporting it evolve.

The Agent 365 SDK - now generally available - allows developers to integrate observability, access controls, and compliance enforcement directly into development workflows. Agents built on any AI platform can be enterprise-ready and compliant from the first line of code. If governance requires a separate step after the fact, it gets skipped. The SDK removes that excuse.

6.  Measure value - not just activity. An agent can be compliant and technically healthy while still failing to create value. Enterprises therefore need to connect governance with business outcomes: adoption, quality, cycle time, cost, risk reduction, employee experience, and realised ROI. Leaders should know which agents deserve expansion, which require redesign, and which should be retired.

Windows 365 for Agents, now generally available, extends runtime governance further: purpose-built Cloud PCs where agents run in fully isolated, policy-governed environments with Entra Conditional Access enforced at the device boundary. For organisations running autonomous agents against sensitive systems, isolated execution environments are no longer optional.

 

From Control Plane to Operating Model

Technology provides the control plane. Organisations still need an operating model - one that defines ownership, governance, risk management, human oversight patterns, incident response, lifecycle, adoption, performance measurement, cost optimisation, and continuous improvement.

Without this operating model, enterprises risk deploying hundreds of AI agents without understanding whether they are secure, compliant, or delivering measurable value. This is the role of AgenticOps - connecting the technical control plane with the management disciplines required to operate an AI workforce continuously, not as a one-time implementation, but as an evolving enterprise capability.

 

 

Model Context Protocol has become the standard integration layer across Microsoft's stack - Foundry, Agent 365, Work IQ, Fabric IQ, Teams, and Copilot all speak it natively. Agents built on any MCP-compatible framework can be grounded in governed enterprise data and composed into M365 workflows without custom integration work. The interoperability that used to require months of professional services is increasingly handled at the protocol layer.

The AI Workforce Lifecycle

Leading organisations are already thinking beyond deployment. The lifecycle of an AI agent increasingly resembles that of an employee:

 

 

Treating agents as long-lived enterprise assets, not temporary experiments, will become a hallmark of mature AI organisations. Each stage demands deliberate process: design with business outcomes in mind, build with governance embedded, register with clear ownership, approve through structured release gates, deploy with monitoring from day one, optimise continuously against measurable outcomes, and retire with the same discipline applied to onboarding.

This lifecycle framing also changes how organisations budget and resource AI. Agents are not one-time project deliverables. They are operational assets that require ongoing stewardship - and that stewardship is where the competitive advantage compounds.

What's Available Today vs. What's Coming

For organisations evaluating Agent 365 now, the question is not whether to wait for the platform to mature - it has materially matured since preview. The more useful question is which capabilities are GA today versus in active preview. Here is the current state as of July 2026:

 

 

The practical implication: you can run a meaningful governance programme today with GA capabilities alone. Preview features represent the hardening of security posture management and local agent discovery - important for organisations with large RPA estates or developer-built agents on managed endpoints.

The Evolution of Enterprise AI

This evolution reflects a broader shift in what enterprise AI actually means - and where the next competitive advantage will be built:

 

 

Agent 365 is the infrastructure for the fourth chapter. The organisations investing in AI Operations discipline now are building the competitive foundation that will be visible in analyst reports two years from now.

The CXO Mandate: Govern Before You Scale

Before expanding AI across the enterprise, leadership teams should be able to answer a small set of consequential questions:

▸  Do we have a reliable inventory of every AI agent operating in our environment - approved and unapproved?

▸  Does every agent have an accountable owner, a defined purpose, and a documented scope of action?

▸  Are identities and data access governed consistently through least privilege and policy - not inherited assumptions?

▸  Can we observe agent interactions, investigate incidents, explain every autonomous decision, and demonstrate compliance?

▸  Do we have lifecycle controls to approve, update, suspend, reassign, and retire agents?

▸  Do our developers have a path to build governed agents by default - or does governance require a separate step that will predictably get skipped?

▸  Can we connect each production agent to adoption, cost, risk, and measurable business outcomes?

▸  Is there a clear operating model spanning IT, security, data, legal, risk, and business ownership?

 

 

The organisations working through these honestly - rather than checking them off on a slide - tend to show up a year later in the analyst reports as Frontier Firms. Not because they moved fastest. Because they didn't end up having to walk anything back.

Broad adoption of agents is happening whether any of us are ready for it. The platform to govern that adoption is now generally available and maturing rapidly. What's still open is whether each organisation chooses to use it before or after they discover the problem.

The Next Competitive Advantage

The first generation of enterprise AI focused on deploying AI. The next generation will focus on operating it.

The enterprises that succeed will not necessarily be those that build the most agents. They will be the organisations that develop the strongest capability to govern, operate, measure, and continuously improve an AI workforce.

Microsoft Agent 365 provides the enterprise control plane that organisations need to govern an AI workforce. The organisations that combine that control plane with a disciplined Enterprise AI Operations model will define the next generation of Frontier Firms.

Every enterprise is building a new workforce. The strategic question is whether that workforce will emerge as a collection of disconnected experiments or as a visible, secure, governed, and continuously optimised source of business value.

 

 

Because the future won't belong to organisations that simply deploy more AI. It will belong to those that operate it better.

 

Selected Microsoft Sources

 

▸  Microsoft Agent 365 overview  —  learn.microsoft.com/en-us/microsoft-agent-365/overview

▸  Microsoft Agent 365: The control plane for agents  —  microsoft.com/en-us/microsoft-agent-365

▸  Microsoft Agent 365 now generally available (May 1, 2026)  —  microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available

▸  What's new in Agent 365 — May 2026  —  techcommunity.microsoft.com/blog/agent-365-blog

▸  What's new in Agent 365 — June 2026  —  techcommunity.microsoft.com/blog/agent-365-blog

▸  Microsoft Build 2026: Securing code, agents, and models  —  microsoft.com/en-us/security/blog/2026/06/02/microsoft-build-2026

▸  Agent 365 SDK overview  —  learn.microsoft.com/en-us/microsoft-agent-365/developer/agent-365-sdk

▸  Microsoft Entra: Governing agent identities  —  learn.microsoft.com/en-us/entra/id-governance/agent-id-governance-overview

▸  Microsoft Entra: What are agent identities?  —  learn.microsoft.com/en-us/entra/agent-id/what-are-agent-identities

▸  2025 Work Trend Index: The Frontier Firm is born  —  blogs.microsoft.com/blog/2025/04/23/the-2025-annual-work-trend-index

▸  2026 Work Trend Index: Agents, human agency, and organisational readiness  —  microsoft.com/en-us/worklab/work-trend-index/agents-human-agency

▸  What's new in observability at Build 2026  —  techcommunity.microsoft.com/blog/azureobservabilityblog