Enterprises bet on AI agents—and many didn’t get what they paid for
Enterprises across Europe and beyond have spent heavily on AI agents—software designed to autonomously complete tasks, coordinate workflows, and interact with internal systems. Yet despite the investment, many early rollouts have produced uneven outcomes, with pilots stalling and promised productivity gains proving difficult to measure.
Why results fell short
Industry observers cite a familiar set of problems. First, many organizations underestimated the data work required to make agents reliable. Without clean permissions, consistent knowledge bases, and well-documented processes, agents can produce brittle or incorrect outputs—especially when asked to execute multi-step actions.
Second, ownership has often been unclear. In some companies, IT, product teams, and business units each launched separate initiatives, leading to duplicated tools, inconsistent standards, and limited reuse of components. Third, governance and risk controls lagged behind adoption. Concerns around security, compliance, and auditability have pushed some deployments back into “experiment” mode rather than production.
What comes next
The next phase is expected to look less like broad experimentation and more like operational engineering. Enterprises are increasingly prioritizing narrow, high-frequency use cases—such as customer support triage, internal knowledge retrieval, and back-office automation—where success metrics can be clearly defined.
Vendors and in-house teams are also moving toward stronger evaluation frameworks, including automated testing for tool use, guardrails for sensitive actions, and monitoring that can trace an agent’s decisions end-to-end. In parallel, budgets are shifting from flashy demos to foundational work: data readiness, system integration, and change management.
For many buyers, the lesson is not that AI agents don’t work—but that scaling them requires the same discipline as any enterprise software rollout, with clearer accountability and measurable outcomes.
Related
