Share with your CTO
Four enterprise technology leaders, from NetApp, Dell Technologies, Birlasoft, and Red Hat, converge on the same uncomfortable claim: the bottleneck in autonomous AI execution is not the model, it’s the stack beneath it. Fragmented data across on-premises systems and multiple clouds, immature edge architectures, and weak governance controls are what’s actually preventing AI from moving from recommendation to action. Birlasoft’s Optimus initiative and Red Hat’s Model Context Protocol extensions are cited as early indicators of what enterprise-grade agentic infrastructure looks like in practice.
What this means for your business
The organizations furthest along in agentic AI, where systems are resolving IT incidents, triggering logistics decisions, or blocking transactions without a human in the loop, didn’t get there by finding a better model. They got there by fixing their data plumbing first. If your enterprise is still treating AI infrastructure as a follow-on concern to model selection, you’re building in the wrong order. The question worth asking is whether your current data environment can be trusted to act on, not just queried for insight.
The governance gap is where most deployments will actually break. Once an AI agent executes rather than recommends, a wrong output isn’t a bad suggestion you can dismiss. It’s an action already taken, possibly inside a production system. The auditing requirement that follows, what the agent did, under what authority, with what data, and who owns the outcome, is a fundamentally different control problem than anything most enterprises have built for. Birlasoft’s framing of “end-to-end traceability” and Red Hat’s push for least-privilege access and revocation controls are pointing at the same gap from different angles. Few enterprises have closed it.
Worth noting that all four contributors here are vendors selling into the infrastructure and services layer they’re describing as essential, which tilts the timeline optimism and conveniently positions their own stacks as the missing piece. That doesn’t make the underlying argument wrong. The dependency they’re describing is real. But the implication that buying the right platform resolves the governance challenge is where the framing oversells. Governance is an organizational capability, not a product feature, and no amount of hybrid multicloud architecture closes the accountability gap if the operating model hasn’t been redesigned around it. The falsification condition is straightforward: if enterprises that deploy purpose-built agentic infrastructure without redesigning human oversight protocols show materially better outcomes than those that don’t, the vendors are right. The evidence for that doesn’t exist yet.
Concept deep-dive: Agentic AI
Agentic AI refers to systems that don’t just generate outputs but take sequences of actions autonomously to complete a goal, like a junior employee who doesn’t wait to be told each next step. Unlike a chatbot that answers a question, an agent might detect a network vulnerability, test a fix in an isolated environment, and deploy the patch, all without human instruction at each stage. The business stakes are high because the same autonomy that creates efficiency also removes the friction that catches mistakes.
Based on reporting from The infrastructure beneath autonomous execution, originally published 2026-06-14 08:19:00.
