Why Most AI Agents Never Leave the Pilot Stage
A survey of 300 executives finds that data alone isn't the problem — context is. Only 34% of agentic projects reach production.
By The energy-climate Desk
Enterprise AI agents have a knowledge problem. According to a survey of 300 data, AI, and other technology executives conducted by MIT Technology Review Insights, only 34% of agentic AI projects make it into production — stalled not by a shortage of data, but by a failure to make that data meaningful. [1]
The distinction matters. Knowledge, as the report describes it, is the understanding of what the data means in the context of individual organizations — the layer that lets agents reason, decide, and act rather than merely retrieve. Without it, agents are prone to making flawed and unreliable decisions. [1]
A subset of firms has cracked this. Production leaders — organizations where, on average, 61% of agentic projects advance beyond pilot — show measurably stronger knowledge capabilities, particularly in semantics. [1] Data fragmentation is the most commonly cited barrier to scaling, flagged by 55% of respondents, while security and privacy concerns weigh more heavily on the leaders already in production (72%). [1]
To close the gap, executives plan to invest in retrieval technologies, like ingestion pipelines, AI-ready APIs, and retrieval-augmented generation (RAG); in AI evaluation agents; and in knowledge graphs — all aimed at building what experts call a knowledge layer between an organization's data and its agents. [1][2]
The pattern is familiar: the infrastructure gap between pilots and production is rarely about ambition, and almost always about plumbing. What agents need isn't more data — it's the architecture to understand what they're looking at.
Legacy data systems, security concerns, and fragmented pipelines are the structural barriers holding agentic AI back. [1] The firms pulling ahead aren't necessarily running smarter models — they've just done the quieter work of giving those models something to reason with.
Without sufficient knowledge, agents are prone to making flawed and unreliable decisions.MIT Technology Review Insights