AI is on nearly every desktop, but it's in almost no workflows. The gap is becoming a structural divide.

McKinsey's State of AI in 2025 report finds that 88% of organizations use AI tools in at least one business function – but only about a third have begun scaling it to the enterprise. Just 23% have moved agentic AI beyond pilots. MIT puts the number of enterprises with AI actually integrated into workflows at only 5%.

Recent research reported in the Harvard Business Review points to the same pattern, of widespread experimentation but limited integration.

Access is not integration, but AI becomes truly valuable only when it is embedded and integrated into processes and data. Companies that are working toward this are the AI Haves. Those that aren’t, the Have-Nots.

The Haves are integrating AI into the core of their workflows: connecting agents to systems of record, standardizing orchestration across teams, instrumenting outcomes at the enterprise level. For them, AI is becoming infrastructure, deliberately, with governance designed in.

The Have-Nots may look similar from the outside. Employees have AI access. They have tools. But in these companies, AI remains an overlay, scattered across browsers, siloed inside departments, detached from governance and workflow architecture. Effective integration is still rare – but it is becoming increasingly critical.

The three markers of an AI-have enterprise are as follows:

1. AI Is Embedded in Core Workflows

In AI-Have enterprises, AI participates in the transaction itself. It does not advise from the sidelines. Agents operate inside systems of record, influencing how records are created, updated, and acted upon.

This is the major shift: from augmentation (early experiments) to integration. Instead of employees copying outputs from chat windows into enterprise systems, intelligence is woven directly into the systems where work already lives. AI analyzes, flags, routes, summarizes, and recommends in context, connected to enterprise data and governed by enterprise rules.

When embedded at this level, decisions are informed at the point of action, not after the fact. Updates occur within operational systems, not in parallel documents or isolated tools.

AI becomes part of how work moves, not an optional layer that employees may or may not use.

2. AI Is Visible and Governed at the Enterprise Layer

AI-Have enterprises do not allow automation to operate in the shadows. AI activity is instrumented, observable, and accountable. Leadership can see where agents are deployed, what data they access, and how outcomes are changing as a result.

The visibility is crucial, not just a monitoring layer. Once AI is embedded in workflows, it becomes operational infrastructure. Infrastructure that cannot be seen cannot be governed, and infrastructure that cannot be governed cannot scale. Attribution matters. Enterprises must be able to trace what model acted, what data it accessed, and why a decision was made.

As AI becomes embedded in core workflows, small, well-intentioned departmental deployments can quietly introduce divergence at the system level. If different systems are powered by different models, trained on different data, or governed by different standards, the enterprise begins to operate on competing versions of reality. Orders route differently. Risk scores diverge. Decisions fragment.

Without enterprise visibility, AI remains fragmented. Enterprise visibility allows AI to mature into infrastructure.

Recent high-profile AI-related failures at companies like Amazon and Meta underscore this point. They show how quickly AI can create risk when it is introduced into existing processes without sufficient oversight and control.

In both cases, the issue was not the presence of AI, but how it was deployed, governed, and monitored in practice. These incidents highlight how quickly AI can create risk when it operates outside of governed, observable workflows.

3. AI Reshapes the Operating Model

The difference between experimentation and integration reveals itself in how the enterprise operates. AI-Have enterprises reorganize how work is initiated, how decisions are made, and how teams coordinate. Processes that were once driven by calendar-based checkpoints and manual escalation begin to respond to real-time signals. Tasks that required human routing and follow-up become orchestrated across systems. Intelligence is embedded within workflow.

This changes more than velocity. It changes structure. Calendar-based processes give way to signal-based ones. Manual coordination gives way to orchestration. In these enterprises, intelligence is no longer purchased seat by seat; it is embedded system by system.

Over time, these shifts compound. Enterprises that integrate AI begin operating as connected systems. Their AI becomes infrastructure.

The AI Have-Nots

AI Have-Not companies are not anti-AI. Many are early adopters. They run pilots, roll out copilots widely, and encourage teams to experiment across the business, for example with low-code tooling for agent creation.

The organizations are pilot-rich, but transformation-poor. They do not integrate AI into the enterprise control plane. In AI Have-Nots, intelligence remains primarily a tool for personal productivity. Agents live in chat experiences or fragmented point tools rather than operating inside systems of record, so the underlying workflows remain manual even when AI offers suggestions.

In practice, this creates a visibility gap: outcomes are not consistently instrumented, traces are not retained in a way that supports audit and learning loops, and governance is reactive rather than designed into the process lifecycle of build, deploy, and monitor.

In competing AI-Have companies, product roadmaps are shifting from using AI single assistant interactions toward multi-step, multi-agent orchestration, with explicit emphasis on evaluation frameworks, quality signals, and operational monitoring. These companies raise the bar for enterprises that are still treating AI as optional assistance, not embedded in execution.

The AI Have-Nots may look innovative on the surface, but structurally they still operate as collections of individuals enhanced by AI, not as coordinated systems shaped by it. In AI-Have enterprises, intelligence moves with the work, inside the systems where permissions, policy, audit, and outcomes already live. In AI Have Nots, intelligence floats around the edges of work, which makes trust difficult: the enterprise cannot reliably explain what the AI did, what it touched, or why.

AI Have-Nots don’t know how to trust AI, and it holds them back.

Closing the AI Gap

Enterprises do not drift into AI maturity. They design for it.

Trust in AI is structural, not cultural. It is built by exposing AI to consequential workflows, instrumenting what it does, and accepting that oversight must evolve alongside automation. It requires moving beyond pilots and beyond productivity boosts, into systems based on observability and accountability loops.

This is uncomfortable work. It means rewriting workflows. Standardizing models. It means hardening APIs and data access. It means deciding which decisions can be automated and which must remain human.

These are not decisions best left to organic adoption; they belong at the enterprise layer.

Without those decisions, AI remains peripheral. With them, AI becomes infrastructure – it can affect how the enterprise sees, thinks, and acts.

We list the best IT documentation tools.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit