London Tech Week Masterclass: why agentic AI needs enterprise control

At London Tech Week, Deliverance AI hosted a Masterclass on one of the biggest questions now facing regulated enterprises: how do you move AI from pilots into production without losing control?

The session brought together Deliverance AI, HPE and NVIDIA to examine what it takes to run agentic AI inside enterprise environments, where data, cost, infrastructure, governance and accountability all matter.

Set against the launch of Deliverance AI from stealth, the discussion focused on a change already underway. Enterprises have spent heavily on GPUs, cloud platforms and AI experimentation. The next challenge is less about access to models and more about operating them safely and economically once agents start making decisions, calling tools and acting across business workflows.

“AI being something that's in production and actually delivering real value. The problem is, who controls it?”

For boards, that question is no longer theoretical. AI is now linked directly to cost reduction, revenue growth and public service outcomes. But many pilot projects stall before production because organisations lack the control layer to manage cost, context, policy and risk once usage grows.

The HPE and NVIDIA perspective added an infrastructure view. Agentic AI has to be designed against the reality of existing IT estates, available energy, data requirements and unit economics. In production, the cost of generating value through AI depends on the relationship between data, energy and infrastructure. That makes control over deployment, workload placement and cost as important as model performance.

Sovereignty was a central theme. Recent events have made the issue real for European and UK boards. When a frontier AI model can be switched off for non-US users by government directive, dependency stops being an abstract concern. For banks, public sector organisations and regulated industries, it becomes a concentration risk built into core workflows.

That is why sovereignty is not only about geography. It is about autonomy. Organisations need to know who controls the model, where the data sits, how agents are governed and who, ultimately, holds the off switch.

This is also why the answer cannot simply be to choose a different model provider. Enterprises need company-specific secure intelligence: open-source or frontier models running inside their own environment, governed by an orchestration layer they control, on infrastructure that cannot be revoked from outside.

That is where Deliverance AI fits.

“One central control plane that observes, manages, reports on, and controls every interaction from a model, from a tool, from a user, from a function perspective.”

The Masterclass made the case for a new operating model for enterprise AI: governed, observable, accountable and built for production from day one.