As companies race to deploy agentic AI, a new consensus is forming around the gap between ambition and readiness. The technology is advancing. The foundations underneath it are not keeping pace. Data is fragmented, infrastructure is built for a different paradigm, and the organizational structures designed for human workers are not designed for agents. Each problem compounds the others.
First, let’s define the problem. CMSWire reported that autonomous AI systems cannot reliably act when enterprise data is isolated across disconnected platforms. The piece argues that most organizations have plenty of data. The problem is that it is trapped in different systems that do not share context. A failed payment in billing and a denied claim in a medical database look like two separate events to an agent with no unified view. CMSWire calls the fix an AI-ready data foundation: a governed, real-time environment that brings operational, customer and interaction data together into a single layer agents can understand and act on.
The article frames data readiness as a competitive differentiator, not just a technical requirement. The question for enterprise leaders is no longer whether they have enough data, but whether their data is ready to work for them. Organizations that treat that question as infrastructure will move faster than those treating it as a project.
MIT Technology Review says that the deployment gap is not just technical. Eighty-five percent of organizations say they want to be agentic within the next three years. Seventy-six percent say their current operations and infrastructure cannot support that change, citing a lack of readiness across people, processes and workflows. The piece argues that many organizations are layering agents onto existing human operating models rather than redesigning how work gets done.
The ROI evidence is pointed. When one enterprise customer switched from activity metrics to outcome metrics, measured return on investment from agentic AI tripled within two quarters. The agents had not changed. The evaluation frame had. MIT Technology Review presents this as evidence that agentic AI requires a systems-level redesign across the technology stack, the workforce and the metrics used to track success. McKinsey predicts that by 2030, three-quarters of current jobs will require redesign, upskilling or redeployment. That timeline is not far off.
The Infrastructure Problem
TechRadar frames that agentic AI demands a fundamentally different compute foundation than the query-driven AI popularized by chatbots and copilots. A single agentic workflow can involve multiple model calls, data retrieval, validation loops and downstream integrations. Unlike episodic inference, agents run continuously. That always-on consumption profile creates cost structures most enterprise infrastructure budgets were not sized for.
Advertisement: Scroll to Continue
TechRadar argues that the economics of agentic AI are defined less by model acquisition than by the ongoing cost of sustained autonomous activity. Energy consumption, cooling requirements, utilization rates, and operational overhead become the dominant variables. Infrastructure optimized for peak performance without regard for steady-state efficiency compounds those costs over time. The piece’s core point is that efficiency, not capability, determines which organizations can actually scale.
The sectors moving fastest are those where the operational payoff is clearest. Healthcare, pharma and financial services are deploying agents because the productivity case is concrete. The infrastructure bill is becoming equally concrete. Organizations preparing for agentic AI by evaluating models first are starting in the wrong place.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
