The Infrastructure AI Agents Actually Need

Recently at E.ON's Tech Summit in Budapest, I presented alongside Dogan and Jose on agentic AI — exploring use cases, organizational readiness, and the foundations we're defining: common terminology, our AI autonomy framework, and the infrastructure required to scale agents responsibly. Sebastian Weber's keynote on an AI-first future set the tone.

The Playmaker Problem

One insight kept surfacing: we don't have a vision problem for AI agents. We have an infrastructure challenge. Enterprises deploy autonomous agents on architectures built for human workflows — not systems that orchestrate across platforms, make real-time decisions, and collaborate autonomously. It's like putting a world-class playmaker on the field when nobody else knows the plays.

Five Non-Negotiable Foundations

From our work, five elements are non-negotiable.

Event-driven architectures for real-time state. Agents at machine speed need to see system changes as they happen. Batch processing creates blind spots; without this, agents decide on stale data.

Semantic layers for unified data access. Agents can't operate autonomously if humans translate between systems. Semantic understanding — knowledge graphs, unified query languages — lets agents discover and access data without hardcoded integrations.

Data synchronization and interoperability across agent networks. In multi-agent systems, consistency is survival. Agents from different vendors must coordinate on shared state and capabilities. Conflicting data means contradictory decisions. We've seen this fail: one agent operating on cached state while another sees live data, executing different actions.

Non-human identity and privileged access management. Agents aren't fast users. They're autonomous executors at machine speed with privileged access. Traditional PAM breaks down. Agents need cryptographic identity, real-time revocation, and continuous monitoring — architected from the start. Zero-trust principles and least privilege are non-negotiable: agents must prove identity per action, not per session.

Governance by design with built-in observability. Compliance added after deployment fails. Agents need transparency, traceability, escalation paths, and real-time visibility embedded from day one. You can't govern what you can't see. Governance is infrastructure.

Where to Start

Most enterprises aren't ready — and that's the point. This foundation is what separates pilots from production. Organizations investing in these building blocks today will be the ones that scale agents beyond demos.

Take one workflow that matters. Instrument it. The infrastructure gaps will show up fast — and the infrastructure you build for that one workflow is the infrastructure your next twenty agents will inherit.