Most engineers remember their internship years. As an Engineer in Training (EIT), a graduate moved from academia into the plant to suddenly be exposed to real processes, complex constraints and the familiar pressure of working with limited time and budget. The EIT period typically lasted three years.
While in training, young process engineers often worked alongside highly experienced operators who knew their plants in intimate detail. These operators were practical, grounded and deeply capable. The operators often scoffed at the junior engineer’s theory, but the relationship still rested on mutual respect. Engineers respected the operators’ experience and intuition, which was on the mark most of the time. Operators in turn respected the confidence of young engineers who walked into a live plant and could apply thermodynamics, mass and heat transfer, and other scientific principles to improve performance. Young process engineers got plenty wrong. Even so, there was enough substance in their contribution for the team to be able to solve hard problems together.
The Context Has Changed
By 2026, many of those experienced operators have now retired. Plants are operating under tighter constraints. Fewer people. Less experience. More production pressure. The system as a whole exists in a tenuous equilibrium. Engineers carry heavier workloads and are expected to do more with less. Projects are being fast-tracked. Time is always short.
To add to this, advanced AI technologies have been introduced into this already fragile environment.
I recently ran an introductory AI workshop for an engineering consultancy made up of seasoned professionals, all from the capital-intensive project world. Before the session, I ran a short survey to gauge the company’s understanding of AI and its level of adoption. The results were revealing. AI use was far from pervasive. A small number of individuals were pushing the boundaries, while most respondents remained uncommitted. Many had tried AI and often failed to get useful results. At best, the output from AI was sporadically helpful.
There were also engineering sceptics who saw no value in AI in their work. They had judged the technology by the slop produced by colleagues over recent months and concluded that AI added no value. In some cases, they believed it created more work, because correcting poor quality AI output took more time and effort than doing the job properly from the start.
Four Principles for Engineers Using AI
The workshop itself went well, though I would have preferred a full day on the subject rather than a few short hours. After preparing the material and getting feedback from workshop delegates, I came back to a set of principles that should help shape an engineer’s view of AI:
- AI should amplify an engineer’s expertise, not replace it.
- Engineering judgement, context and experience is hard won and will never be fully replaced by software algorithms.
- Human expertise remains central to complex problem solving.
- Engineers must always own the recommendation.
That third point matters. AI performs well on certain categories of problems, but engineers know that problems in heavy industrial plants rarely come down to physics and chemistry alone. The people who design, operate and optimise a plant form a complex ecosystem. An AI tool is very unlikely to accurately model human interactions and their impacts on the physical plant with any certainty.
The fourth point matters just as much. Handing engineering responsibility over to a large language model is a very bad idea. The trap is easy to fall into because these tools can produce fluent, polished output that looks professional and well grounded.
The Real Risk: Cognitive Surrender
The cognitive implications of continued AI use in engineering work are serious. Under time pressure, engineers and experienced operators alike can hand over critical thinking to a tool, with disastrous consequences. This does not come from carelessness. It comes from a subtle convergence of factors: polished and fluent AI output, time pressure, and hidden assumptions and hallucinations embedded in the AI response. Researchers are already studying this cognitive surrender, and a growing body of scientifically sound work is raising clear warning flags.
Where AI Already Helps
Most engineers will meet modern AI tools first through personal productivity. Tools such as Claude, ChatGPT and Gemini are already being used in day-to-day work to draft reports, write e-mail replies, summarise documents and handle similar tasks. In a second project, I am working with specialists who are using AI for their project development methodologies, and the AI tools are proving invaluable for assessing document quality during the pre-feasibility and FEED phases, at key decision gates and helping to efficiently reach FID and beyond.
Modern AI tools can check documents for inconsistencies, summarise large volumes of information, compare documents against required standards and support several core value assurance and document review tasks. Used well, they make a project engineer extremely productive. However, the risk of cognitive surrender still sits in the background. In project environments that move quickly and carry a high degree of complexity, people are already under stress. AI output is compelling. Accepting it without question is sometimes too easy.
How to Work With AI Properly
Our Engineers and AI workshop covered prompting techniques, but one principle stood above the rest: work with AI as a collaborator. That means using prompts that force the tool to expose hidden assumptions, uncertainties and risks in its own response. A critical engineer does not accept the first answer from an AI tool. They follow up with challenges such as:
- “List every assumption embedded in your response”
- “What are the three biggest weaknesses in your answer”
These follow-up prompts often reveal far more value than expected. By investing another 30 seconds, an engineer can usually reveal deeper insights that help to build confidence in the depth and integrity of the final result.
Final Thought
Engineers once had to learn how to work effectively with experienced operators. They now need to learn how to work with AI in the same collaborative spirit. AI will have a significant impact on how work gets done in the future, and that applies across all engineering disciplines. After our workshop, I came away convinced of the enormous potential; but that engineers need to be intentional about building at least a basic level of AI literacy so they can start taking full advantage of the available technology.
