Monday morning: why Microsoft and Google are targeting the AI workflow layer
Two companies spent twenty years building the substrate of how enterprise work gets done. They are now converting that asset into an AI workflow franchise. Most of the market doesn’t see it.
Something important happened in the last three weeks, and it received far less attention than it deserved. The companies that already own how enterprise work gets done just activated that ownership — and the window for anyone else to contest the broad market is closing faster than most enterprise technology leaders have recognized.
Over the last few weeks, two things happened that deserve more attention than they’ve received. On March 9, Microsoft shipped Wave 3 of Microsoft 365 Copilot — not as a preview or a pilot, but as a production release embedding agentic capabilities directly into Word, Excel, PowerPoint, and Outlook. On March 26, Google launched an agent step inside Workspace’s Opal platform: autonomous, goal-seeking workflows with persistent memory and dynamic routing, operating entirely within the data environment enterprises already use and govern. Both shipped quietly, without the fanfare that typically accompanies a major AI announcement. That quiet is worth noticing.
The enterprise AI conversation has been dominated, understandably, by the rapid improvement of foundation models and the proliferating ecosystem of standalone AI agents that run on top of them. Those things matter. But they have drawn attention away from a structural shift that may prove more durable: Microsoft and Google are in the process of converting two decades of enterprise infrastructure investment into an AI workflow franchise. The mechanism is straightforward. The implications are significant.
THE DATA ASSET HIDING IN PLAIN SIGHT
The pattern should be familiar to anyone who has followed how AI outcomes are determined: the model is rarely the decisive variable. The data underneath it — its depth, its governance, its proximity to how work happens — is what separates durable advantage from temporary capability. Nowhere is that more visible than in the enterprise workflow layer. Think about what an enterprise generates in the course of a working week. Decisions get made — in email threads, in meeting rooms, in comments on shared documents. Relationships form and evolve — tracked in calendar patterns, in who gets cc’d, in who reviews whose work. Context accumulates — in the slide decks that capture a strategy, in the contract negotiations that establish a precedent, in the project post-mortems that explain why something went wrong. This is not data in the conventional sense. It is the organizational memory of how work is done.
For most large enterprises, this memory lives almost entirely inside two platforms: Microsoft 365 and Google Workspace. Not by design, exactly — but as the accumulated consequence of two decades of deployment, standardization, and organizational habit. The email is in Outlook or Gmail. The documents are in SharePoint or Drive. The meetings are in Teams or Meet. The org chart is known; more importantly, the work chart — who collaborates with whom — is knowable, because the evidence is all there.
Microsoft and Google don’t need to build a data moat. They’ve been sitting in one for twenty years. What changed is that they now have the agentic capability to use it.
This is the asset. And until recently, it was largely latent — useful for search, for compliance, for basic automation, but not for the kind of contextual, multi-step reasoning that makes AI genuinely transformative in a workflow context. The arrival of capable foundation models changed that equation. Microsoft and Google didn’t need to go acquire organizational data. They already had it, in quantity, with permissions, governance, and trust already established. What they needed was the AI capability layer to activate it. That layer has now arrived.
WHAT ACTIVATION LOOKS LIKE
Microsoft’s Wave 3 centers on a concept called Work IQ — an intelligence layer that grounds Copilot agents not in generic knowledge but in the specific patterns of your organization. Who you collaborate with. What you work on. How decisions get made, as evidenced by the email trails and document histories that accumulate in M365. Agents built on this foundation don’t need to be told the context. They inherit it. They can draft and send emails, schedule meetings, create documents, and execute multi-step workflows without ever leaving the data environment where the work already lives — and without requiring the IT integration project that a standalone agent deployment would demand.
The governance architecture is equally important, and equally underappreciated. The permissions model, audit trails, data residency compliance, and security policies that Copilot agents operate under are not new constructs. They are the same governance infrastructure that enterprises spent years and significant budget building around M365. Every agent Microsoft ships inherits that infrastructure by default. For a CIO evaluating enterprise AI deployment at scale, this is not a marginal advantage. It may be the deciding factor.
Google’s architecture is structurally parallel. Workspace Studio’s no-code agent builder puts agent creation within reach of non-technical employees — grounded in Gmail, Drive, Docs, and Meet. Gemini Enterprise extends reach into broader enterprise systems. The Agent2Agent (A2A) protocol, co-developed with Salesforce, establishes an open interoperability standard that allows agents across platforms to hand off work — a move that simultaneously expands the Workspace ecosystem and positions Google as a standards-setter rather than a silo. The Opal agent step, launched this week, adds autonomous goal-seeking logic, persistent memory across sessions, and dynamic routing: the full agentic stack, available without leaving the environment employees already work in.
The governance model comes pre-installed, because it was already there. That is not a feature. That is a twenty-year head start.
Both companies are also investing in multi-agent orchestration — frameworks that allow specialized agents to collaborate within a governed environment, handing off tasks between each other as work moves across functions. This is the architecture of a genuinely new model of enterprise software: not a tool that individuals use, but a participant layer woven into how organizations operate. The productivity suite becomes the operating system for AI-enabled work.
WHY THE INTEGRATION TAX MATTERS
The counterargument — that standalone agents offer superior model capability and greater flexibility — is true to a point. The most capable reasoning available today does not ship natively inside M365 or Workspace. For technically sophisticated practitioners — developers, data engineers, specialized operators — a well-configured standalone agent can outperform what either platform delivers. That segment of the market is real and will remain contested.
But for the vast majority of the enterprise workforce — the finance manager, the HR business partner, the regional sales lead, the operations director — the relevant question is not which agent scores highest on a reasoning benchmark. It is which agent can actually be deployed, governed, and trusted at scale without a six-month integration project. Connecting a standalone agent to the full context of how an organization works requires solving, from scratch, exactly the problems that Microsoft and Google have already solved: data access, permission modeling, governance infrastructure, security architecture, and the organizational trust that comes from years of enterprise deployment.
Standalone agents are already in the enterprise market, and some deployments are genuinely impressive. But the operational reality for most organizations is that ease of use and permissions are not secondary concerns — they are the primary gating factors for deployment at scale. A standalone agent built outside of an application, however capable, arrives without the permission model, the governance infrastructure, or the organizational trust that enterprise IT requires before granting an autonomous system access to sensitive data and workflows. Building that foundation from scratch is a significant project. The ClawHavoc incident in February — in which attackers flooded OpenClaw’s ClawHub marketplace with malicious skills by exploiting a marketplace that required nothing more than a week-old GitHub account to publish — illustrated concretely what happens when agentic systems scale without that infrastructure in place. Microsoft and Google don’t face that problem. Their agents inherit permissions, governance, and trust that already exist.
THE FRANCHISE QUESTION
What Microsoft and Google are building is in the most precise sense, a franchise. The underlying models are not proprietary — both companies license frontier model capability and will continue to do so. The data is not proprietary in the traditional sense — it belongs to the enterprises that generated it. What is proprietary, and what compounds over time, is the combination: the grounding layer that connects model capability to organizational context, the governance architecture that makes enterprise-scale deployment viable, and the distribution advantage of being the platform that employees already open every morning.
This is a structurally different competitive position from what either company held in previous technology cycles. In cloud, in search, in mobile, the competitive dynamic was ultimately about platform capability. In enterprise AI workflows, the competitive dynamic is about who already has permission to act inside an organization’s most sensitive data environment — and has the governance architecture to do so responsibly. That permission was not granted last quarter. It was granted over years of enterprise deployment, security certification, compliance investment, and organizational trust-building.
The AI workflow opportunity is large. The number of companies structurally positioned to capture the broad enterprise market — not the developer segment, not the greenfield deployment, but the Monday-morning knowledge worker at scale — is two.
The number deserves scrutiny, because there are serious candidates that complicate it — and they deserve more than a footnote. The right frame is not “who else is doing AI” but “who else holds a structural data asset that is native to how enterprise work gets done, at scale, with governance already established.” That is a high bar. A handful of companies clear it, each in a different functional domain.
Salesforce is the most prominent. Through Slack, it owns a meaningful layer of the organizational data stack — the conversational record of how work gets decided, the channel discussions, the cross-functional threads that capture context before it makes it into any formal document. In many enterprises, Slack is where much of the real work happens: where decisions get made, where priorities shift, where the informal org chart becomes visible. Combined with Salesforce’s CRM depth — the customer relationship data, the pipeline history, the account context that M365 and Workspace do not natively hold — the Salesforce AI play has a genuine data moat of its own, one directly relevant to the workflows that drive revenue. Agentforce is a serious product from a company that understands enterprise deployment. It should not be dismissed.
ServiceNow owns substantial IT and operational workflow record — tickets, incidents, change requests, and approval chains across the enterprise. Where M365 and Workspace capture how people collaborate and communicate, ServiceNow captures how work gets resolved: the execution layer, not just the conversation layer. For large organizations, that data asset is deeply embedded, highly governed, and directly relevant to a category of AI workflows — IT operations, service management, cross-functional approvals — where autonomous agents can deliver measurable value quickly. Now Assist, its agentic platform, reflects a company moving to unlock new value for customers with AI.
Workday holds the HR and finance workflow record — headcount decisions, compensation history, budget approvals, workforce planning data accumulated over years of deployment. That data is sensitive, embedded, and relevant to a significant category of enterprise AI use cases that neither M365 nor Workspace natively serves. SAP occupies a similar position in supply chain, procurement, and manufacturing workflows, particularly in industrial enterprises where its penetration is deep and the data asset is genuinely irreplaceable.
What each of these companies shares is a partial moat — deep in a specific functional domain, but not broad across the full knowledge-work substrate. None of them, individually, holds the horizontal data coverage that M365 and Workspace hold across the full working week of so many enterprise employees. Whether any can credibly expand beyond their functional stronghold — or whether the right outcome is a multi-agent world where functional specialists and horizontal platforms interoperate — is the genuinely open question. What is clear is that the structural contest for the horizontal, organization-wide, knowledge-worker AI workflow market has a defined set of serious contenders — and the companies that entered 2026 with decades of organizational data, pre-installed governance, and a workforce that opens their platform first thing every morning are not in a weak position.
WHAT THIS MEANS
Enterprise technology moves in long cycles. The shift from on-premise to cloud took the better part of a decade to fully play out. The shift from licensed software to SaaS took longer. The shift to AI-enabled workflows will be faster — the underlying models are improving too quickly, and the competitive pressure too acute, for the usual enterprise adoption timeline to hold. But it will not be instantaneous, and the outcome is not yet fully determined.
What is already determined is the structural advantage. Two companies entered this cycle with assets that took twenty years to build and cannot be replicated quickly: the organizational data, the governance architecture, the enterprise trust, and the distribution of being the platform where work already happens. They are now activating those assets with agentic AI capability that, as of this month, is no longer on a roadmap. It is in production.
The model race will continue. It will generate headlines, benchmarks, and genuine capability improvements that matter at the margin. But the workflow layer — the layer where AI stops being a tool you consult and starts being a participant in how your organization operates — is being claimed right now, quietly, by the companies that already own Monday morning.
The question for every enterprise technology leader isn’t which model to buy. It’s whether they’ve thought carefully enough about which platform is about to become the operating system for how their organization works.
Crawford Del Prete is a Senior Advisor at PSG Equity and former President of IDC. He writes about enterprise software, AI, and technology market dynamics. All opinions are his.