IBM urges built-in controls as agentic AI expands

At IBM's Think on Tour EMEA 2026 conference in London, executive Ana Paula Assis highlighted a stark gap between enterprise expectations and operational readiness. IBM research reveals that 80% of surveyed executives anticipate AI driving revenue growth, yet a mere 20% can clearly articulate how that will occur. Furthermore, 80% of organizations reported that an AI-related issue has reached their boardroom or C-suite, while three-quarters of EMEA executives expect autonomous agents to execute full business tasks and make decisions without human intervention by 2027. To help address these management challenges, IBM introduced Sovereign Core, an open-source platform designed to grant organizations control over data, infrastructure, and model coordination.

Industry experts at the event emphasized that oversight must be engineered directly into infrastructure rather than added later. Michael Nimhauser, global head of sovereignty at Atos, noted that corporate anxiety over losing control extends across data storage, underlying hardware, and agent control systems. U.K. managed services provider Celerity, led by CEO Craig Aston, uses IBM's AI-powered software development tool, Bob, alongside a dual-review process that balances innovation with strict assessments of risk, data safety, and access permissions.

Practical case studies show how human oversight remains essential during automation. U.K. insurer Markerstudy deployed Bob to evaluate a core legacy software system containing six million lines of code built over roughly 30 years. The tool completed a comprehensive four-week assessment, generating a risk heat map to guide targeted modernization instead of a complete system replacement. Markerstudy CTO Dan Fiehn explained that keeping engineering teams actively validating output transformed a subjective evaluation into an objective decision-making process.

For IT managers and enterprise architects, these developments signal a fundamental transition in how software is governed. As agents move beyond basic queries into executing workflows, technology leaders must institute parallel approval structures that measure potential return on investment against operational exposure before rolling out autonomous capabilities.