Every operator, engineer, and reliability lead is being asked some version of the same question at the moment: what does AI actually mean for manufacturing jobs? A recent Plant Engineering article on the topic, drawing on Rockwell Automation’s tenth annual State of Smart Manufacturing Report, lands in a relatively narrow band of the public debate. AI is not coming for industrial jobs in the way the headlines have suggested. It is changing the shape of the work, the skills that earn a premium, and the daily operating model of asset-intensive plants. For organisations running enterprise asset management programmes, this matters more than another wave of vendor announcements, because the operating reality on a maintenance shift is where these claims get tested first.
This piece looks at what the research is saying, what it implies for asset management and maintenance roles specifically, and where the gap between the technology and the operating model is still wide.
What The Research Actually Says
Rockwell’s tenth annual State of Smart Manufacturing Report, published in June 2025 and based on a survey of 1,560 manufacturers across 17 countries, is one of the larger recent data sets on industrial AI adoption. A few numbers from the report are worth pulling out, because they shape the rest of the conversation.
- 95% of manufacturers have invested in, or plan to invest in, AI or machine learning over the next five years.
- 48% of manufacturers expect to repurpose or hire additional workers as a result of smart manufacturing investments, and 41% are using AI and automation specifically to help close skills gaps and address labour shortages.
- The share of respondents calling the ability to apply AI an “extremely important” skill jumped from 10% to roughly 47% in twelve months.
- Quality control remains the most common AI use case for a second year, with 50% planning to apply AI or ML to product quality in 2025.
- Cybersecurity is now the second largest external risk respondents identify, with 49% planning to use AI for cybersecurity in 2025, up from 40% the year before.
The headline that gets reported, that AI is reshaping rather than eliminating manufacturing jobs, is supported by the underlying data. The signals around hiring, skills demand, and the labour shortage being a top barrier to competitiveness all point in the same direction. The plants that will struggle in the next few years are not the ones with too many people; they are the ones that cannot find or develop the people they need.
From Automation To Industrial Autonomy
The framing the research uses, picked up by Plant Engineering, is that the move underway is from automation to industrial autonomy. The distinction matters. Automation tends to handle known, deterministic tasks against a defined recipe. Autonomous systems sense, predict, act, and adapt against operational goals, with humans setting intent, boundaries, and exception handling rather than every step.
For asset management this is not a hypothetical shift. It is already visible in three areas:
- Predictive and condition-based maintenance moving from scheduled task lists toward continuous health monitoring, with technicians prioritising the alerts the model surfaces rather than working a calendar of routines.
- Energy and utilities optimisation handed to closed-loop systems, with engineers reviewing anomalies and tuning the operating envelope rather than collecting and plotting data themselves.
- Quality and visual inspection workflows handled by trained models, with quality engineers triaging the edge cases that the models flag rather than running every check by hand.
In each case the work has not disappeared. It has moved up the decision hierarchy.
The Asset Management Consequence
For organisations running an EAM platform such as IBM Maximo or MAS, this evolution lands directly on roles that have existed for years: maintenance planner, reliability engineer, asset performance analyst, maintenance technician. None of those jobs are going away. What changes is the proportion of the day spent on tactical execution versus supervision and strategy.
A maintenance planner whose system can suggest the next likely failures, propose work order templates against them, and check parts availability automatically spends less time assembling work packs and more time challenging the suggestions, sequencing them against operational constraints, and managing exceptions. A reliability engineer with anomaly detection running across the historian spends less time scanning trends and more time investigating the cases where the model is right, or where it is wrong in an informative way.
That is a more demanding job, not an easier one. It requires comfort with data, model behaviour, and the limits of probabilistic outputs that the existing workforce does not uniformly have. The sharp jump in respondents calling AI competence extremely important is the labour market expressing that demand in real time.
Where Job Design Is Lagging The Technology
The risk in industrial AI rollouts is not that the technology will fail. It is that the operating model will not catch up. A few patterns are showing up consistently in plants that have made early investments:
- Job descriptions still reward tactical throughput. If a planner is measured on the number of work orders scheduled, the AI tools that should free them to do higher-value work get treated as another tab to ignore.
- The handover between supervisory and autonomous tasks is informal. Who decides when the model is trusted to act unsupervised? Who logs the override? Without that discipline, governance breaks down quickly.
- Skills development is treated as a training catalogue rather than a structured programme. AI literacy for asset teams is not a one-day course. It is a curriculum that sits across data fluency, model behaviour, statistics, and domain reasoning.
- The data foundation underneath the model is rarely audited honestly. Models do not perform well on top of inconsistent failure coding, missing asset hierarchies, and untagged sensor streams, and most operators know which of those problems they have.
These are operating model questions, not technology questions. They are also the gap between the studies showing 95% AI investment and the small number of plants that can credibly claim to have changed the work.
Practical Priorities For Asset-Intensive Organisations
For organisations running asset-intensive operations, a few priorities are worth treating seriously over the next twelve months.
- Audit the current state of the asset data underneath the EAM platform before scaling any AI use case. Failure coding, asset hierarchy, sensor metadata, and work order completion quality matter more than the choice of model.
- Rebuild the job descriptions of planners, reliability engineers, and maintenance supervisors against the operating model the AI tooling actually creates. Reward judgement and exception handling, not raw volume.
- Plan a structured AI literacy programme for the asset and reliability functions. Treat it the same way the safety case for a new piece of plant would be treated: as a serious, audited capability, not an awareness video.
- Set explicit policies for when models can act autonomously, when they advise a human, and how overrides are recorded. This is the only governance discipline that holds up at scale.
- Pair every AI initiative with a workforce plan that identifies who is being upskilled, who is being repurposed, and which roles are genuinely new. The headlines about repurposing only mean something if there is an internal programme behind them.
The organisations that get this right will end up with smaller, more capable teams managing larger and more reliable estates. The organisations that ignore it will run AI pilots that never make it into the operating model.
The Bigger Picture
The trend reflected in the recent research is consistent across the credible studies in this space. Industrial AI is moving from analytics layered on top of existing systems into the systems themselves, and the jobs adjacent to those systems are changing accordingly. The story of AI in manufacturing jobs is not one of displacement. It is one of a labour market that already cannot fill its open roles, a generation of plants that need to operate at higher complexity than the previous one, and a set of decisions about how to design the work.
Asset management leaders sit at the centre of that conversation. The technology decisions will be made anyway. How the work is organised, measured, and developed is what will determine whether the investment pays back, and those decisions are made inside the operating model rather than inside the AI platform.