Enterprise Agentic AI Adoption Is Still Below 1 Out Of 10

Errol Gardner puts the current state of enterprise agentic AI adoption at “less than 1” on a 0-to-10 scale. He is not saying organizations are unaware of AI, or that they are ignoring it. His point is narrower and more consequential: production-grade agentic AI, deployed across meaningful workflows in large enterprises, remains very early.

Gardner distinguishes among three layers that are often collapsed into one market narrative. Traditional AI and machine learning are already embedded at scale in many organizations, across processes and industries, and have been driving efficiencies for years. Generative AI, particularly large language models used to process information and produce text-like responses, has also spread widely. Organizations are using public information, internet-scale tools, and private models built inside corporate firewalls to support research, analysis, knowledge access, and internal productivity.

Agentic AI is different. In Gardner’s framing, the transformational claim begins when agents replace activities, those activities join into workflows, workflows become business processes, and processes form part of a larger value chain. That is not a tool adoption problem. It is a redesign of how the organization works.

Craig Smith tests the adoption question by comparing agentic AI to cloud. Smith suggests cloud might be around 7 out of 10 in enterprise adoption: most organizations have embraced it and moved what they can. Gardner accepts the comparison but challenges the implied maturity. Perhaps seven out of ten organizations have embraced cloud, he says, but the real question is how widely they use it and what they decide can be moved. Even within the same industry, different companies will make different judgments about what should or should not move to cloud.

The comparison matters because cloud has been around for roughly 15 years and still has not reached full penetration. Gardner is not arguing that agentic AI will necessarily take 15 years to reach maturity. He is arguing against the opposite claim: that agentic adoption across enterprises will happen in “15 days or even 15 months.” In his view, cloud may itself be a prerequisite for many agentic deployments, which creates a further dependency.

The current state he describes is uneven. In Gardner’s rough view, a relatively large number of organizations may be using agentic systems “a little bit” in production — he suggests perhaps around 20% — but only in very small parts of the business. That is different from broad deployment across workflows and value chains. Many enterprises can have agentic pilots, internal demos, and isolated production use while still being nowhere near agentic transformation.

Gardner’s “less than 1” is also not meant as pessimism about the technology’s eventual importance. He presents it as a measure of remaining opportunity. If the technology can be deployed safely and responsibly, it could create new ways for employees to work and new sources of business value. But that is different from treating agentic systems as already broadly deployed across enterprise operations.

The possibility that trust and reliability problems could confine agents to narrow tasks does not lead Gardner to dismiss the agentic thesis. He expects more organizations to move from experimentation to adoption at scale in certain elements of their business and parts of their value chains. But he frames the path from experiments to scaled workflows as a difficult hurdle. A single useful agent is not the same as an enterprise-grade agentic workflow embedded in accountability, controls, data access, management behavior, regulatory context, and the expectations of the people whose work is being changed.

For Gardner, the single biggest impediment to AI adoption is usually “something related to the human beings.” That includes leaders, sponsors, middle managers, workers, or some combination of them. The technology may be new, but the adoption problem is familiar: large organizations are hard to change.

Gardner argues that anyone who treats large-scale organizational change as easy has either not done it before or is being “slightly economical” about what is achievable. Large enterprises are not simply slow because they lack imagination. They have structures, management systems, regulatory obligations, established work styles, and operating mechanisms that helped make them successful. Those same mechanisms can become barriers when the company tries to change quickly.

The point applies especially to agentic AI. A single employee using a tool for email or analysis is not the same as a company rebuilding a process around agents. A production workflow implicates data governance, customer information, employee information, regulatory duties, accountability, and the willingness of managers and employees to let work be performed differently.

Gardner sees a gap between the market’s language and what large organizations can operationalize. The benefits being discussed — productivity, efficiency, better customer experience, new forms of value creation — may be real. But the route to them passes through the existing organization. It requires capabilities that scale not just across one task, but across a full process, across a value chain, across geographies, and across functions.

The adoption challenge also creates tension between bottom-up experimentation and top-down control. Employees may be comfortable using generative AI tools in their private lives and may want to use the same tools at work. That creates risk if corporate data, customer data, employee data, or proprietary information is moved into public systems. Gardner says major organizations are putting in guardrails to prevent that from happening and monitoring movement of data from work laptops into uncontrolled platforms.

But guardrails can feel like constraints. A creative employee may be told to innovate with AI while being restricted to the organization’s approved model, approved interface, or private corporate LLM. The company wants AI literacy and productivity gains; the employee may want flexibility and access to the tools they already prefer. Gardner sees that disconnect as a real adoption issue, not a side concern.

That is why he does not treat agentic deployment as a pure software rollout. Even where the technical system works, the business still has to align sponsors, leaders, managers, and workers around new patterns of work. If workers believe enterprise-grade AI will displace them, resistance is not irrational; it is predictable.

Errol Gardner says EY cannot credibly advise clients on AI transformation unless it is deploying AI inside its own business. EY is a large enough organization for that internal work to resemble the enterprise adoption problem it advises on: roughly 400,000 people globally, across consulting, tax, audit, and other areas.

EY made an early decision to give as many employees as possible access to generative AI tools within its own environment. It built a private LLM, moved internal knowledge and information into it, and gave employees controlled access so they could experiment with corporate data without pushing it into uncontrolled public systems. That deployment was paired with learning and development, especially for employees without native technology skills.

Gardner’s emphasis is on lowering the psychological barrier to use. For nontechnical professionals, generative AI can make technology feel as accessible as a search engine. The objective was to move employees away from “I’m not a technologist, I can’t do this” and toward experimentation with internal tools, internal knowledge, and approved workflows.

The internal use cases Gardner names are practical rather than futuristic. EY has used AI to improve research, client proposal work, contracting, consolidation of information, and access to knowledge distributed across the global organization. The goal is to let a consultant sitting anywhere retrieve relevant knowledge faster and bring it to a client more effectively.

EY also has about 90,000 technologists in the business, according to Gardner, helping build and engineer more sophisticated solutions. The firm works with third-party ecosystem partners as well. But Gardner frames the broad deployment less as a technology build alone than as an organizational campaign: giving employees tools, giving them training, asking them to experiment, and pushing the message that EY must “show up differently” to be credible in the market.

On agentic AI specifically, Gardner says EY has been more focused for roughly the last 12 months. The firm has an “ever-increasing number of agents” inside the business. At first, he says, counting them became “a bit of a sport,” but because EY is large and spread across many countries, counting agents became less useful than assessing them qualitatively.

The agentic use cases he mentions include email-related tasks, information analysis, updating employees on new information hourly or daily, and service-delivery tasks with consistent patterns. He points to tax work, including producing returns; audit delivery; and the software development lifecycle, where agentic approaches are driving efficiencies.

Gardner does not present EY as already fully transformed by agents. He describes the work as maturing and still subject to the same scaling challenge that other corporate organizations face. But the internal deployment matters because it changes how EY sells and delivers consulting. It also forces the firm to confront the same questions its clients face: which processes can be changed, which data can be used, which controls are required, and how quickly people can adapt.

AI is changing consulting, and Gardner says EY builds systems as well as advises. But he situates that change within the consulting industry’s broader history. Consulting is, in his words, fundamentally about helping organizations change, so consulting firms have had to change with each major shift in the working world.

He cites the movement of work offshore, including the growth of large practices in India and elsewhere, and the rise of in-house consulting teams inside corporations. AI is another shift in that sequence, not an isolated event.

The most immediate change is in service delivery. Before 2020, Gardner says, consulting was an industry where many clients believed work was happening only if consultants were physically present in the client’s office. COVID changed that expectation around presence. AI is changing another expectation: how consulting work should be measured.

Historically, consulting was often measured by inputs: hours worked, days worked, and the effort required to produce a deliverable. AI makes that model harder to sustain. If a client assumes AI was used, the client may reasonably ask why the work took as long or cost as much as before. Gardner sees both challenge and opportunity in that shift. AI can let consultants do work faster, but it can also let them do more and achieve more than they could historically.

That creates pressure toward outputs rather than inputs. The client cares less about the visible labor that produced a deliverable and more about the outcome and business value. Gardner does not say the consulting industry has fully solved this commercial model shift. He presents it as one of the ways AI is actively changing how consulting operates.

Most client demand, however, is still about traditional business problems. Companies want to launch products, acquire organizations, deepen customer penetration, improve customer experience, drive productivity, increase supply-chain efficiency, and solve other familiar business issues. What has changed is that clients increasingly want AI used as part of the answer. They are not typically asking, in Gardner’s words, for “an AI thing” or for someone to “agentify” the business without context. They are asking how to achieve traditional business outcomes faster, more effectively, or more efficiently with AI.

That is especially visible in long-term technology projects. A company replacing systems of record or changing its ERP platform might be undertaking a three-year project that is not, in itself, an AI project. But clients increasingly ask whether it can be done faster, and whether AI can be deployed to do it differently than it would have been done in the past. They may even build in expected savings from future AI development over the life of the project.

Craig Smith raises a broader convergence: consulting firms are building systems; software companies are acting like consultants by auditing client workflows and identifying automation opportunities; LLMs have ingested public consulting reports and can produce advice; and startups may include consulting-like services to land products inside enterprises. Smith’s framing is that consulting, software, and AI seem to be “morphing into a new industry.”

Gardner accepts that the ecosystem is changing and that organizational boundaries are blurring at the margins. Corporations, technology companies, startups, service providers, and consulting firms are all expanding or rethinking what they do. He suggests the likely “sweet spot” is collaboration among multiple actors rather than any one category fully absorbing the others.

But he resists the idea that consulting can simply be codified into a product. Implementing technology inside a business to generate a business outcome remains, in his view, a service discipline grounded in experience with change. As long as organizations are composed of human beings trying to produce outcomes through other human beings, they will want to interact with people in that change process.

Even if a future workforce blends agents and people, the human side remains a possible limiter. Gardner’s argument is not that products cannot automate advisory tasks or embed expertise. It is that the hard part of enterprise change includes human alignment, resistance, governance, and practical implementation — and those are not yet reducible to a product in the way some projections assume.

He is skeptical of a “touchless” vision of corporate operation with no humans in the loop. He treats it as an interesting theoretical model, then asks how it would work in practice. For Gardner, technology exists to improve human and corporate life, not to make people redundant as the central objective.

The “less than 1” adoption thesis is not only about internal corporate friction. Gardner also points to macro constraints that could slow AI deployment even where the technology improves. The adoption path for agentic AI runs through governments, labor markets, national data policy, and the distribution of control over critical infrastructure.

One constraint is data sovereignty. Gardner contrasts the current environment with the early 2010s, when globalization was more often assumed to be an enduring direction of travel. Today, countries are more sensitive about data, information, critical infrastructure, and dependence on technologies delivered from outside their borders.

He sees sovereignty concerns especially in Europe, while noting that similar dynamics exist in the United States and many other countries. Governments do not want transformative technologies to become dependencies over which they have little control if something goes wrong. That can slow enterprise AI deployment, particularly for critical infrastructure and heavily regulated industries.

A second constraint is workforce displacement. If AI systems can displace work at scale, Gardner asks who will act as custodian to protect employees from mass redundancy. Is that a company’s role? The technology firms’ role? Or will governments have to do it? Gardner’s view is that governments are ultimately likely to have to lean in.

The mechanisms he names are broad possibilities rather than a policy program: regulation, peer pressure, taxes, or other ways of disincentivizing organizations from replacing people with robots, agents, or comparable systems at scale. His point is that large-scale labor displacement would become a public-policy issue, and that public-policy response could put brakes on some technologies or on how quickly companies adopt them.

Craig Smith broadly agrees that workforce transition will need to be addressed by governments, though he speculates that taxation to fund universal basic income may be more plausible than direct regulation of hiring and firing. Gardner’s broader claim is more general: the displacement issue is too large to be treated as a narrow company-by-company implementation matter.

At the same time, Gardner identifies a countervailing pressure: many Western countries face declining working-age populations. A shrinking productive workforce is increasingly responsible for supporting both the young and the old. AI could help close that productive-capacity gap. The difficulty is finding a path where AI expands capacity without producing mass displacement.

Gardner says there is no model, platform, or empirical evidence that cleanly solves that balance. That absence makes government involvement more likely in his view. It also complicates any simple prediction that enterprises will adopt agentic AI as quickly as the technology becomes available.

Gardner raises a further issue that is not mainly about adoption inside one company: the distribution of value between technology firms and traditional industries. AI is creating a large value uplift for the technology sector, not only in stock-market terms but in the broader transfer of workload and effort from traditional companies to technology providers.

If consumer companies, oil and gas companies, financial services firms, and other traditional businesses increasingly rely on technology firms to perform more work on their behalf, Gardner asks how sustainable those business models are over the longer term. The question is not simply whether AI vendors will make money. It is whether the balance of power and economic value between sectors remains viable.

His concern is that the technology sector could become so powerful that it controls many other sectors, or even replaces parts of the services currently offered by major organizations. He does not present this as a settled outcome. He frames it as a dynamic to watch: where is the equity in the relationship, and how can that balance be maintained?

That point connects to his preference for collaboration over category collapse. Enterprises, consultants, technology companies, and startups may all be necessary to create compelling solutions. But the long-term value exchange between them remains unresolved.

Agentic-first startups will emerge across industries and compete with traditional leaders, Gardner says. His simple answer is “yes, no doubt.” But the stronger claim requires sector-by-sector and value-chain-by-value-chain analysis.

Agentic models lend themselves most naturally to industries centered on information processing, where the product is data, digital output, or a service rather than a physical good. Financial services is Gardner’s obvious example. But even there, the same sector characteristics that make agentic AI attractive can also make adoption hard. Financial services is heavily regulated; a startup cannot simply enter and take over parts of the business without meeting the regulatory framework.

Gardner expects startups to take over specific activities or parts of value chains that might previously have been outsourced. In yesterday’s model, a company might outsource a function; in tomorrow’s model, a startup might provide an agentic service that effectively removes that work from the incumbent organization.

Scaling that from a narrow wedge into broad enterprise disruption is harder. Gardner distinguishes between consumer adoption and enterprise adoption. Technology startups operating in a business-to-consumer market can sometimes find ways to reach consumers quickly. Selling into enterprises is very different. “Getting through to an enterprise buyer is not straightforward,” he says. That makes it hard for startups to scale across multiple companies and geographies, even if their product is compelling.

This is one of the recurring tensions in Gardner’s view. Agentic AI may be powerful enough to create new companies and threaten incumbents, but the enterprise environment slows, filters, and reshapes adoption. The technical possibility of automation is not the same as the commercial ability to sell and implement it across regulated, global organizations.

Gardner says EY’s Workforce Reimagined work points to anxiety, learning gaps, and the need for clearer leadership communication. Employees are reading amplified claims about AI online and on social media, and many are worried about what those claims mean for their roles. That anxiety can become resistance when organizations try to deploy enterprise-wide AI systems.

The issue is not only fear of job loss. Gardner says there is also a gap in the learning and development employees receive to use AI tools effectively in their work environment. Companies may encourage adoption, but employees need training, permission, and practical support to use AI in ways aligned with the organization’s controls.

More importantly, leaders need to articulate a vision. Gardner says employees can become worried that by adopting and using AI in day-to-day work, they are being moved toward an unclear end state. They may not know whether the organization’s philosophy is augmentation, efficiency, headcount reduction, reinvention, or some mix of these. Without clarity from the C-suite and executive sponsors, adoption campaigns can feel threatening.

The leadership task, in Gardner’s view, is to communicate how the organization wants to embrace AI, what that means for people, and what guardrails will shape deployment. That communication needs to be broader than instructions for individual roles. It should explain the organization’s philosophy and the intended relationship between AI, employees, and future work.

Gardner also identifies an intergenerational management problem. Younger workers may be more willing to experiment with tools and innovate, while their managers may come from a generation less inclined to use the tools personally. Those managers may not necessarily encourage or reward employees who are pushing AI adoption.

That intergenerational divide can slow adoption even when the company formally supports AI use. A junior employee may see the productivity upside; a manager may not know how to evaluate the work or may not reward the experimentation. In Gardner’s account, workforce adoption is therefore not simply a matter of giving employees access. It requires communication, guardrails, and management that does not become a brake on the experimentation the organization says it wants.