Scaling AI agents is where things get complicated

It’s no secret companies are moving beyond pilots with AI agents and beginning to deploy them across business functions. But new research from Deloitte suggests the bigger challenge isn’t getting agents into the enterprise, it’s preparing the organization to work with them.

Nearly two-thirds of executives surveyed by Deloitte say their organizations are reevaluating their business models because of advances in agentic AI. Yet fewer than half say their organizations are prepared for agentic AI across the six areas of the business examined in the study. The report finds business processes and workforce readiness are the weakest areas.

The findings come from a survey of 501 U.S. senior managers and C-suite executives directly involved in their organizations’ agentic AI strategy or implementation. All of the organizations surveyed were at least piloting agentic AI. Deloitte also interviewed 20 executives and AI and data-science leaders at large organizations across industries.

The research suggests that companies are progressing through the early stages of agentic adoption at different speeds. Forty-two percent of respondents said their organizations are still testing small numbers of agents or have a few deployments underway. Another 43% said they are expanding deployment across functions. Just 15% said they have reached scaled, orchestrated, multi-agent adoption.

Where scaling is occurring, Deloitte says organizations are concentrating primarily on lower-risk, repeatable use cases with near-term returns, including applications in engineering, customer service and IT.

The challenge is that moving from individual applications to an enterprise built around autonomous systems requires a heck of a lot more than deploying yet another layer of technology.

The report’s authors say an agentic enterprise is likely to operate fundamentally differently from today’s organizations. That means organizations will need to redesign the work and the associated systems around AI agents rather than simply inserting agents into existing processes.

This transition is proving difficult.

Only about half of the executives surveyed say they have a clear view of their future operating model with AI agents. Three factors are identified as the biggest obstacles to scaling: 72% cite a lack of a unified and accessible data foundation, 70% cite an inability to trust and govern agents, and 67% cite the cost and complexity of integration.

The weaknesses become particularly apparent when companies look at the processes that agents are supposed to perform.

Only 16% of respondents say their business processes are prepared for agentic adoption, while just 5% say they are highly prepared. Even among organizations that have already reached scaled, orchestrated, multi-agent deployment, only 46% say their business processes are prepared. Just one in five executives says their organization is prepared to redesign processes to run autonomously with AI agents.

Deloitte points to several reasons. Organizations often lack well-documented processes, have fragmented data and systems, and remain constrained by established ways of working. Limited AI fluency among leaders and employees also makes it harder to fundamentally rethink workflows.

For now, many organizations are taking an incremental approach.

Rather than redesigning processes from the ground up, they are layering AI agents on top of existing processes. Deloitte says that approach can make economic sense because it can deliver shorter payback periods and help organizations demonstrate returns on their AI investments.

One healthcare company’s AI architect (who was interviewed but unnamed) described the choice in financial terms: redesigning processes is currently too expensive when the easiest way to demonstrate ROI is to make incremental improvements.

Deloitte does not characterize this approach of layering as inherently wrong. The approach can build operational capacity and organizational credibility while revealing where processes, data and workforce practices need to change.

But layering alone is unlikely to deliver the more fundamental transformation required to become a successful agentic enterprise.

Only 31% of respondents expect at least half of their processes to be redesigned or rebuilt around AI agents within two years. That figure rises to 74% within four years.

A similar pattern emerges when organizations look beyond individual processes. Only 25% expect cross-functional agent coordination within two years, compared with 58% over the following four years.

The workforce is a significant gap

Forty-three percent of executives expect their organizations’ adoption of AI agents to cause “a lot” to “extreme” job disruption within the next 12 to 18 months. That rises to 72% when executives look two to three years ahead.

Yet half of those surveyed say their organizations aren’t investing enough in the workforce transformation required to support agentic AI.

Many companies already provide basic training. Seventy-one percent report baseline AI-agent literacy programs, while 65% report targeted upskilling or reskilling for roles that could be affected by agents. But Deloitte says workers will need to develop capabilities that go beyond basic AI literacy.

As agents become more capable, employees may increasingly shift from executing individual tasks to supervising, validating and orchestrating AI systems. Three-quarters of executives surveyed believe human-agent collaboration will generate more value than pure automation, but fewer than half say their organizations have defined the human-agent operating models needed to achieve that value.

That operating model will require companies to establish which decisions agents can make, when humans must intervene, who remains accountable for outcomes and what expertise should remain human-owned.

Deloitte’s prescription is therefore broader than simply deploying more agents. It recommends that organizations develop an integrated roadmap connecting business outcomes with AI use cases, data, governance, architecture, work design and workforce changes.

It also recommends treating incremental deployment as a bridge to deeper process redesign rather than the final destination, investing more heavily in workforce transformation, and explicitly defining the division of responsibilities between humans and agents.

The underlying message is that agentic AI may eventually change not only individual jobs or workflows, but the way companies organize work itself.

Most organizations surveyed have moved beyond the proof-of-concept stage, Deloitte says, but scaled orchestration remains the exception. The larger group is still layering agents onto existing processes in pursuit of quick wins.

The next stage of the transition will require those companies to decide what they want the organization to look like when AI agents are no longer simply tools employees use, but participants in how work gets done.

As one life sciences executive interviewed by Deloitte put it, the goal is not simply to have more agents. An agentic enterprise is one that embeds fit-for-purpose agents into its core operations, with autonomous systems making decisions and humans providing oversight where it matters.

For companies moving toward that model, the technology will ultimately be only one part of the transformation. The harder question may be whether the organization itself is ready to make the change.

The post Scaling AI agents is where things get complicated appeared first on Engineering.com.