IBM Warns Enterprise AI Agents Demand New Operating Models as Governance Pressures Mount
At IBM's Think on Tour EMEA conference, executives warned that the rush to deploy autonomous AI agents is exposing enterprises to operational risks, urging companies to embed governance by design.
As artificial intelligence evolves from answering queries to executing multi-step tasks and making business decisions, enterprise leaders face an escalating accountability gap. At IBM’s Think on Tour EMEA 2026 conference in London, technology executives and industry partners gathered to address the urgent need to balance rapid automation with strict operational oversight.
Ana Paula Assis, senior vice president and chair for EMEA and APAC at IBM, highlighted the disconnect between corporate ambitions and operational readiness. According to IBM data presented at the event, 80% of executives expect AI to deliver revenue growth, yet only 20% can articulate precisely how that growth will materialize.
Furthermore, 80% of surveyed EMEA executives reported that at least one AI-related issue had already escalated to the boardroom or C-suite. Despite these friction points, adoption is accelerating toward greater autonomy: three-quarters of executives surveyed across EMEA expect AI agents to execute complete business tasks and make decisions without human intervention by 2027.
“The question now is: Can we operate it at scale safely, reliably?” Assis said in her keynote. “That is not a technology problem. That is an operating problem, and it requires a fundamentally different, new AI operating model.”
Governance Cannot Be an Afterthought
Industry leaders on stage emphasized that deploying autonomous agents requires maintaining control over models, underlying data, and core infrastructure.
Michael Nimhauser, global head of sovereignty at French IT services and consulting firm Atos, observed that while companies are drawn to the potential of AI, anxiety over losing operational control remains widespread. Nimhauser stressed that control mechanisms cannot be tacked on at the end of a project. Instead, governance must be built into system architectures from the design phase.
Craig Aston, CEO of U.K.-based managed services provider Celerity, echoed the need for balance. Celerity, an IBM partner that uses IBM’s AI-powered software development tool Bob internally, conducts parallel reviews for every AI initiative—evaluating business value and return on investment on one track, while independently assessing data security, access permissions, and risk on the other.
Aston noted that while companies should foster internal innovation, enthusiasm must not bypass organizational safeguards, warning that unmanaged risks could threaten a company's survival.
Keeping Humans in the Loop
Organizations putting AI agents into practice are finding that human supervision remains indispensable. Markerstudy, a U.K.-based insurer, deployed IBM's Bob tool to modernize core legacy systems built over approximately 30 years.
Bob analyzed six million lines of legacy code in four weeks, generating a heat map that identified critical components needing updates. This enabled Markerstudy to carry out targeted modernization rather than an expensive, full-system replacement.
Markerstudy CTO Dan Fiehn stressed that the deployment was far from autonomous. Human developers remained actively involved to validate outputs and train models throughout the analysis. Fiehn emphasized that the primary goal should not simply be deploying as much AI as possible, but delivering measured outcomes that enhance decision-making for customers and staff.
Shifting Toward Sovereign and Integrated Architectures
To address growing demands for oversight, IBM introduced Sovereign Core, an open-source platform designed to help organizations retain greater control over how their AI environments are deployed and operated.
IBM executives argued that AI implementations consistently stumble when data remains isolated in silos and IT infrastructure is fragmented. As Assis concluded, AI is shifting from a back-office support mechanism into the foundation of enterprise operations, requiring organizations to rethink how human workflows, decisions, and algorithmic systems interact safely at scale.
