{{item.videoDuration}}
{{item.title}}
{{item.text}}
{{item.videoDuration}}
{{item.text}}
Enterprise evolution follows recognizable patterns. AI is compressing the time between innovation and architectural reckoning.
Key takeways
Every major tech wave follows the same pattern: capability arrives first, and infrastructure catches up later. AI agents are repeating this, but faster.
Enterprises today sit in one of two modes: automating existing tasks or expanding what agents can do.
Adoption moves through four stages: scattered pilots, agent proliferation, a modernization reckoning, and finally the intelligent enterprise.
The real turning point is Stage 3: agents need a place to reason from, a unified data layer, not just another management tool bolted onto the old one.
Companies that build this foundation now reach the intelligent enterprise faster and with less disruption than those who wait for the reckoning to force it.
Executive Summary
Each of the major technology cycles follow the same arc. A new capability arrives first, and the infrastructure catches up later. The gap between the two is where enterprises lose time, money, and competitive ground, adopting in silos against an inconsistent foundation, until they finally get there just as the next cycle begins.
AI agents are the latest instance of this pattern, and the fastest-moving by a wide margin. Frontier model generations that used to arrive annually are now arriving in weeks. Enterprises are already deploying agents in pilots, across individual business units, and in productivity tools embedded across the organization. The early results show limited but real value, and very high potential. That potential and its limits rest on the infrastructure beneath these agents, one built for humans, not for the machines now operating alongside them. When agents begin to scale and interact, connecting the way humans do and in concert with them, operating continuously and acting on enterprise data in real time, the limitations surface. Business processes and systems weren't built for this.
An intelligent enterprise is one where human and digital systems work seamlessly together, in concert across data and insights. Getting there is no easy feat. It requires the next significant wave of enterprise transformation, focused not on new applications or added AI features, but on the substrate beneath them: the unified data, model, and governance layer that AI reasons on top of.
The coming reckonings in this next arc are not a new phenomenon. Cloud's reckoning was about infrastructure, cost, and scale. Mobile's was about the balance of access and security. AI's is about something deeper: the substrate beneath the applications and what reasons over it.
The enterprises that recognize this pattern now and make foundational shifts with the holistic arc in mind can reach the intelligent enterprise faster, with less disruption, and with a compounding advantage over those that don't.
This briefing maps that arc. It is written for leaders who are already investing in AI and want to understand what comes next.
The Pattern We've Seen Before
The history of enterprise technology is not one of the smooth adoptions. It is a history of waves, each one following a recognizable sequence that, in hindsight, looks inevitable and, in the moment, rarely does. Each of the companies sits on its own beach watching the waves arrive, choosing whether to wade in.
The pattern is consistent: a new wave arrives and moves the company forward before the infrastructure exists to support it at scale. Governance and controls come in late, following the proliferation. Somewhere between early adoption and full maturity, there is a reckoning where the gap between what the technology promises and what the company's foundation can support becomes impossible to ignore.
With AI, this reckoning will likely happen again across the business in different ways, unless there is a cohesive, enterprise-wide approach.
Cloud adoption is the clearest parallel for enterprise leaders today. In the early years, cloud adoption was driven by speed and cost, as 'lift and shift' drove workloads to the cloud quickly. With this new scale in hand, IT provisioned infrastructure in hours rather than months, and the business moved faster for it. But many of today’s enterprises didn't re-architect for cloud. They moved existing applications and data structures into new environments without rethinking the underlying design. The result was a slow decade of cloud enhancement work: rationalizing and containing sprawl, redesigning data architectures, rebuilding security models, and learning, often at great expense, that cloud's holistic value required a new foundation purpose-built for it. The capability and potential were real from the beginning. The infrastructure and governance took years to properly manifest.
Mobile followed the same arc in the previous era. Devices proliferated faster than enterprise security, device management, or application architecture could keep pace. The reckoning came in the form of shadow IT, data exposure, and a generation of application re-platforming efforts that lasted well into the next cycle. In this era, the expression "no one wants to end up on the front page of the Wall Street Journal" was born.
The pattern isn't unique to any technology. It is a structural feature of how enterprise adoption works. Companies move toward capability because the early value is genuine, and the potential seems attainable, and perhaps a little competitive fear is at play. The foundational gap only becomes visible at scale. By the time it does, the cost of closing it is significantly higher than it would have been at the start, and the outcomes may vary greatly from the initial promise.
AI and agents are following this arc now, faster than any wave before them. Frontier model generations that used to arrive annually are now arriving in weeks. Agent capabilities are compounded across model releases. The call for improved cross-functional reasoning is too great to ignore, and the reckoning will arrive at the same accelerated pace.
Companies cannot afford to stay on the beach. This wave is moving faster than the last.
Where Enterprises Are Now
Many of today’s enterprises are operating in one of two modes when it comes to AI agents: process enhancements or capability expansion.
Process optimization looks like deploying agents to automate repeatable, well-defined tasks like summarizing documents, routing requests, generating drafts, or accelerating workflows that already exist. It may extend to AI bolted onto existing enterprise software stacks. The value is simple, yet real, and measurable. Companies take business processes they already have and effectively digitize them with new AI tools.
Capability expansion is broader. Agents get embedded into business unit workflows, connected to specific data sources, and interact with enterprise software in ways that go beyond simple task completion. Here too, the early results are genuine.
Both modes have value, but each operates on infrastructure and applications that were never designed for AI.
The platforms governing these deployments, such as security controls, identity management, and compliance frameworks, are largely in place for a majority of enterprises. The cloud-era control plane is already the gate through which production AI operates. This is what enables the early stages of the intelligent enterprise: process optimization and capability expansion.
That governed layer exists, is working, and is even improving as AI progresses. Underneath these controls are more layers that weren't built for AI: the applications that store and surface enterprise data, and the compute infrastructure beneath them. These were designed for human access patterns: structured interfaces, request-response interactions, workflows initiated and supervised by people. Agents don't work that way. They query continuously, connect laterally, can forge new processes, and they demand data availability and system responsiveness at a scale and frequency that human-pattern infrastructure and enterprise software stacks weren't designed to support.
At today’s stage of AI adoption, these limitations are largely invisible, which is precisely what makes this arc challenging.
Enterprises are moving quickly from pilot to siloed AI implementations, and in some cases achieving genuinely integrated results by doing something they could have done all along: connecting data and technology in the collaborative way people already work. Those wins are real, and they matter. But they are also masking a structural constraint that becomes visible the moment someone asks the right question: "Why can't this agent access that dataset?" or "Why does this workflow break when it crosses into that application?"
Unlike a traditional enterprise application, an agent doesn't stabilize after deployment. The appetite around it grows. Teams want to connect more data. Leaders want it to be integrated with more applications. What began as a contained pilot becomes a request for enterprise-wide access, and that is precisely the moment the foundation gap surfaces, and a new data substrate is required.
Without the underlying modernization work completed with the intention to transform the business, progress stalls in the same familiar place: The realization that the company is still stuck in the previous arc.
The four visible stages of AI growth
Stage 1 – Bespoke LLM Projects
When companies start out with AI, it can be a bit of the Wild West. An ambitious employee begins experimenting with an LLM quietly, or a vendor lands a meeting with an interested leader. However, when it begins, the early pattern is consistent: identify a simple use case, solve a pain point, and show something that works.
These projects are typically business unit-led, occasionally IT-adjacent, and rarely enterprise-coordinated. This is how genuine experimentation works. The use cases are narrow enough to deliver real value without requiring deep system integration, and the energy behind them is exactly what early adoption needs.
At this stage, the foundation limitation is genuinely invisible. Deployments touch few systems and make few demands on the infrastructure beneath. Governance is present and manageable across data handling, model access, and output accountability.
The risk lies in what is not yet being thought about. Projects architected at this stage without the governed enterprise platform in mind often cost more to bring into compliance later. The goal is not to slow experimentation, but to confirm that what gets built now does not become the retrofit problem in later stages.
Technologies in play: At this stage, Microsoft entry points are typically uncoordinated and exploratory across Microsoft 365 Copilot for productivity, Copilot Studio for business-unit experimentation, and Microsoft Foundry for developer access to frontier models including GPT and Claude. Each delivers value independently, and each represents a future integration question the substrate will need to answer.
Stage 2 — Agent Proliferation and a Governance Reckoning
The agents proliferate quickly. It turns out there was more than one experiment, or the first one created inspiration elsewhere. Multiple business units are now solving their own problems, often independently, often with different models, different platforms, and different approaches. What began as isolated experimentation starts to look like fragmentation.
This is where the first governance reckoning arrives. Agents can't talk to each other. Workflows stop at application boundaries. IT is asked to be the hero for environments it didn't design and wasn't consulted on. The foundation limitation, invisible at Stage 1, is making itself known.
IT's instinct at this stage is to reach for a management layer that brings agents into a governance framework regardless of which vendor built them or which platform they run on. The instinct is correct, but the tools alone are not sufficient if the architecture beneath them was not designed for what agents can eventually demand of it. Control is necessary, but it is not the end state.
This is where the substrate question first surfaces, usually framed as something narrower, like "how do we get these agents to share context?" or "why does this workflow lose state when it crosses platforms?" The architectural decisions made here go beyond governance and start applying to the substrate decisions in the next stage. The foundation limitation is no longer invisible. It is manageable, but only if the right questions are being asked.
Technologies in play: Microsoft's Agent 365 brings agents into a unified governance plane across identity, lifecycle, observability, and policy applied across agents the way Entra and Intune are applied across users and devices. Microsoft Foundry, with Foundry Agent Service moving into general availability, provides the runtime where agents execute under that governance.
Stage 3 — The Modernization Reckoning
The substrate question that surfaced in Stage 2 cannot be answered with another management layer. By the time an enterprise reaches Stage 3, it is asking something different: where do agents reason from?
This is the modernization reckoning, and at its core, it is a substrate reckoning. Agents should have a place to reason from: a unified, governed data layer where enterprise context lives, accessible to machines and humans alike. This is a new data substrate, emerging as one of the more valuable intangibles the company can create. Without it, agent projects often rebuild the same plumbing, badly, in isolation.
The modernization requirements are specific. Agents connecting to enterprise systems need governed, secured pathways, which means mapping what those systems do, how they relate to each other, and where accountability sits. Application data should move into environments that are machine-accessible, because even mature enterprise applications were not built for the kind of abstraction-layer access agents require. Cloud infrastructure designed around human access patterns, and APIs built for legacy digital access, should be refactored for the way AI will use them.
This is the same modernization cycle enterprises have navigated before. Cloud was this shape, mobile was this shape, but here, it is accelerated by data availability and governed through the platform layer. IT's ability to move at a responsible but quickened pace is the variable that determines trajectory. The model quality, the agent's sophistication, and the ambition of the use case will not matter if the infrastructure layer is not ready.
Technologies in play: Microsoft Fabric and OneLake provide the unified data layer with one governed copy of the enterprise's structured and unstructured data, accessible to humans and agents under the same identity and policy. Microsoft IQ provides the semantic layer that turns raw data into reasoning-ready context: Fabric IQ for business semantics, Work IQ for the workplace, Foundry IQ as the unified retrieval endpoint for agents. Entra ID and Purview extend identity and governance from the user surface down into the substrate itself. Azure provides the underlying infrastructure refactored for machine access patterns. PwC's modernization services bridge where enterprises are today and where the platform should be.
Stage 4 — The Intelligent Enterprise
When the substrate is in place and agents can reason across it, the enterprise itself begins to operate differently.
Enterprise applications stop being the center of gravity. Workflows and processes stop ending at system boundaries. The CDO, the CIO, and the business unit leaders find themselves working from the same architecture diagram with the same imperative, asking the same question: what should our agents be reasoning about next?
This is an intelligent enterprise. It is the operating model that emerges when governed access, governed boundaries, and governed accountability extend across every piece of data and systems, and when agents work the way people do, across functions and in concert with human collaborators rather than alongside them.
A companion briefing examines this state in deeper architectural terms as the shift from systems of record to systems of reasoning. The substrate is its foundation. The agents are its operators. The platform layer is its governor. The intelligent enterprise is what they produce together and what allows the company to finally deliver on the decade-old promise of digital transformation.
PwC is operating in a version of this state internally, with hundreds of agents deployed across tax, assurance, and advisory functions and growing. It is achievable. The question is no longer whether the intelligent enterprise is real, but how quickly the foundation beneath it gets built and which enterprises start building now.
Technologies in play: Microsoft Fabric and Microsoft IQ supply the unified data and semantic context. Microsoft Foundry, with hosted agents now generally available, provides the runtime where agents reason and act under per-agent identity, isolation, and audit. Agent 365 again governs the agent population at scale, the way Entra and Intune govern people and devices, now operating across a mature substrate rather than an emerging one. M365 Copilot becomes the human surface where the substrate's reasoning meets the daily flow of work. PwC's scaffolding, useful in the prior stages, here moves from enabling agent workflows to actively forging new processes across the business as the substrate matures.
A commitment to what’s coming
The enterprises that see the holistic arc today are the ones positioned to make the foundational decisions that Stage 3 and Stage 4 demand. They can reach the intelligent enterprise faster, with less disruption, and with a compounding competitive advantage. The ones that don't will likely face a modernization reckoning that costs more and disrupts more than it needed to.
Unlike previous cycles, there is no time to play catch up. The wave is moving faster than cloud, faster than mobile, and the foundation work should begin before the reckoning forces it. The enterprises building that foundation now are not just preparing for the next arc — they are forging it.
From cloud to AI, only faster
See the four stages toward an intelligent enterprise
{{filterContent.facetedTitle}}
{{contentList.loadingText}}
{{contentList.loadingText}}
Follow us
