AI Autonomy Paradox: Enterprises Restrict AI Execution

Enterprises are increasingly integrating artificial intelligence across their operations, yet a significant structural pattern has emerged: a universal reluctance to grant AI agents full autonomy. A recent PYMNTS Intelligence report, “Wholesale Writes the AI Playbook: How Goods Firms Are Scaling Intelligence Across the Enterprise,” underscores this paradox, revealing that while AI’s analytical capabilities are embraced, its ability to execute actions independently remains severely restricted.

The study, based on a May survey of 60 senior technology executives from U.S. enterprises generating at least $1 billion in annual revenue, found that every wholesale firm, 90% of retailers, and 85% of construction firms confine AI agents to ‘look-up access’ only. This means AI can retrieve information but is explicitly prevented from independent execution. Crucially, no surveyed company across any sub-industry currently permits fully autonomous AI action, establishing a clear boundary in current deployment strategies.

Despite these autonomy limitations, AI deployment is robust and expanding. Wholesale firms, identified as the most mature AI adopters in the goods economy, leverage AI across 35 of the 75 tracked tasks, achieving majority use in 6 out of 8 business functions. Key applications include contract and proposal generation, and security monitoring and threat detection, both adopted by 75% of these firms. Retail and construction sectors also integrate AI extensively, covering approximately 31 tasks each, with a primary focus on marketing and sales functions.

This widespread restriction to look-up access is not a reflection of AI’s technical capabilities, but rather a deliberate governance decision driven by inherent risks. Enterprises deploying AI agents at scale grapple with concerns such as agents operating without traceable identities, acting outside approved parameters, or inadvertently exposing sensitive data. These formidable governance challenges provide the structural explanation for the caution surrounding full execution rights for AI.

Looking ahead, the ambition among firms leans heavily towards semi-autonomous AI with robust human oversight. Within the next five years, 85% of wholesale firms, 60% of retailers, and 55% of construction and manufacturing firms aim for this level of AI integration as their primary goal. Significantly, not a single goods firm across any sub-industry envisions fully autonomous AI as its core future vision. This consistent ceiling implies that while AI will increasingly assist in action, human intervention will remain the ultimate arbiter.

The persistent gap between current look-up-only postures and future semi-autonomous visions highlights a fundamental business process challenge, rather than a mere technological hurdle. Trust, rather than AI model limitations, emerges as the paramount barrier to deeper AI adoption. Previous PYMNTS Intelligence research reinforces this, noting that 98% of product leaders are unprepared to grant core-system access to fully autonomous agents, even as autonomous agents demonstrably automate more accounts receivable work, underscoring the enduring human imperative for control.