AI agents require process redesign and workforce preparation

Summary

AI-agent adoption is moving from experimentation to expansion, but many organizations are still unprepared to change how work gets done. That is the central conclusion of a Deloitte study based on a survey of 501 US leaders and interviews with 20 executives and AI and data leaders.

According to the findings, 43% of organizations are expanding agents across multiple functions and 15% have reached orchestrated multi-agent systems across several workflows. Most organizations, however, are still layering agents onto existing processes instead of redesigning those processes from the ground up.

In practice

Deloitte identifies three main barriers to scale: an insufficient unified and accessible data foundation, cited by 72% of leaders; difficulty trusting and governing agents, cited by 70%; and the cost and complexity of integration, cited by 67%. Fewer than half of respondents consider their organizations prepared for agentic AI across most of the areas examined.

Processes are among the weakest areas: only 16% say their processes are prepared and 5% say they are highly prepared for agent adoption. Many companies are therefore starting with a layering approach, adding agents to existing processes to achieve faster gains. Deloitte sees this as a useful bridge, but not enough if it does not evolve into end-to-end process redesign and cross-functional coordination.

Context

The study also points to significant changes in work. Almost half of leaders expect a lot or extreme job disruption over the next 12 to 18 months; that share rises to 72% over a two- to three-year horizon. Routine and structured tasks are the clearest candidates for autonomy, while creative, strategic and high-judgment work is expected to continue requiring human oversight.

Workforce preparation is not keeping pace with that expectation. Although 71% of organizations are working on baseline AI-agent literacy and 65% are developing upskilling or reskilling efforts for affected roles, half of leaders say their organizations are not investing enough in workforce transformation. The survey was conducted from April to June 2026 among US organizations already piloting agentic AI solutions, so its results do not necessarily represent all companies or other markets.

Deloitte’s recommendation rests on four areas: create an integrated roadmap tied to business outcomes, use incremental deployment as a bridge to redesign, provide teams with time and safe environments for experimentation, and define a human-agent operating model. That model should clarify which tasks can be delegated, who validates outputs, when human intervention is required and who is accountable for decisions.

Why it matters

• Scaling agents depends as much on process quality, data and governance as on model capability.
• Adding agents to fragile processes can speed up an operation without fixing its structure or clarifying accountability.
• Training will need to move from general AI awareness to supervision, validation, orchestration and the ability to challenge outputs.
• The figures describe a trend among leaders already testing agents, not an independent forecast or a universal measure of the market.