A 62-point gap separates manufacturing's AI adoption from its automation at scale

Ten percent of respondents in Parsec's survey run AI or ML enabled automation at scale, and 72% say they use AI somewhere in operations ^[3][1]. Read as shares of the same base, that puts roughly one adopter in seven at scale and leaves 62 percentage points between using AI and running it at scale ^[2][3].

Analytics deployments are more common than automation, and use cases touching production and operational control are absent from the top set ^[5]. Most of the 19-point rise since 2024 landed in applications that read data ^[1][5].

Bill Rokos, who has led development of Parsec's manufacturing operations management platform TrakSYS since 1999, argues the block is psychological as much as operational ^[12]. Generative AI is the most commonly deployed technology in the sector and is probabilistic by design, which he says can make it a poor fit for tasks that require strictly deterministic behavior ^[9]. Where systems are deployed so that their recommendations are hard to interrogate, operators are asked to approve outputs they do not fully understand ^[14]. "Blaming the system or agent only gets you so far, and, to many, it doesn't feel like enough," Rokos wrote ^[13]. He names the candidates for the blame that follows: the operator who approved the output, the vendor, and the CTO who signed off ^[11].

The accountability reading is his, not the respondents'. "Though few respondents in our survey name the phenomenon directly, anxiety about the ambiguity of accountability permeates the findings," Rokos wrote ^[8]. What respondents did rank were governance, cost and integration, which have displaced infrastructure limits at the top of the barrier list ^[4]. Governance is the closest measured proxy for ownership in that list, and it sits beside cost and integration ^[4].

A vendor of manufacturing operations management software benefits when leaders conclude their AI needs a system of record with named approvers. The column also does not describe the survey's methodology, including sample size and field dates ^[15]. Neither of those touches the measured findings. Governance ranks as a barrier, production control is not in the top use cases, and in a plant, per Rokos, a small error can hurt a worker, stop a line or ruin material ^[4][5][10].

One figure is characterized more gently than it reads. Rokos describes risk perception as spreading near evenly between being too hesitant (60%) and too aggressive (40%) ^[6]. That is a 20-point gap ^[4]. The reported majority worry is caution, and leaders are now less enthusiastic about AI than their staff, a reversal of the early-adoption pattern ^[7].

The decision in front of an operations leader this quarter is narrower than accountability in general: which classes of task a probabilistic system is allowed to control, and who signs the approval when it recommends one ^[9][11]. That decision is cheap to defer while the deployment is a dashboard. The use cases where a recommendation moves equipment sit outside the survey's top set ^[5], and that is where deferring gets expensive.