Teacher-centred or student-centred engineering education? And, to complicate matters further in the twenty-first century, should we ban artificial intelligence from the classroom or embrace it blindly as the solution to every educational problem? The algorithm-driven world of social media encourages us to choose a side. Real engineering education should not.

In contemporary engineering education, particularly in strategically important fields such as thermal sciences, renewable energy, and the energy transition, these approaches are often presented as competing ideologies. On one side, we have direct instruction: rigorous explanations, derivations on the board, and carefully structured guidance from the lecturer. On the other, project-based learning, collaboration, student autonomy, and unrestricted access to digital tools, including artificial intelligence. But educating the engineers who will have to address the energy and climate challenges of the twenty-first century requires considerably more judgement than choosing a pedagogical label or declaring war on technology.

There are moments when an engineering student needs an expert to organise knowledge, model a procedure, and make visible the physical intuition that the student has not yet developed. This is particularly important when introducing abstract concepts or material with a high cognitive load: the Second Law of Thermodynamics, transient heat transfer, fluid mechanics governing equations, or the foundations of exergy analysis. Asking students to discover such concepts entirely on their own can easily lead to fragmented understanding. Asking an AI system to explain them without sufficient prior knowledge can be even more problematic, because students may lack the conceptual tools needed to recognise an incorrect or misleading explanation. Clear teacher-led instruction provides structure. It establishes a logical sequence, reduces unnecessary demands on working memory, and ensures that students understand fundamental principles such as conservation laws, energy balances, physical assumptions, and mathematical relationships before they are asked to tackle open-ended engineering problems. Good explanation still matters. Being able to explain the physics behind a phenomenon remains one of the most important things an engineering lecturer can do.

The problem is that engineering education cannot stop there. Solving a standard textbook exercise, particularly one in which the student simply identifies the appropriate equation and substitutes numerical values, provides increasingly limited evidence of engineering competence. An AI system can solve a thermal balance, estimate a cycle efficiency, or generate a piece of MATLAB or Python code in seconds. The important question is therefore no longer simply whether students can obtain an answer. It is whether they can formulate the problem, judge the assumptions, interrogate the result, and defend the engineering decision.

Once students possess the necessary theoretical foundations and mathematical tools, responsibility should gradually shift towards them. This is where artificial intelligence can become something more interesting than a potential source of plagiarism: it can become a strategic tool, a computational assistant, a design copilot, and, in some contexts, a research partner. Students can use AI to explore design alternatives, process large datasets, develop preliminary models, generate code, test scenarios, or identify possible approaches to complex engineering problems. They can work on thermal energy storage systems, renewable microgrids, concentrating solar power plants, or decarbonisation strategies. But the important word is use. The educational objective is not to produce students who know how to operate an AI system. It is to produce engineers who know when they can trust it, when they cannot, and how to find out.

A student who simply presses Run in a simulation package has not necessarily demonstrated understanding. Nor has a student who asks a language model to produce a complete piece of code. Nor has a student who accepts an apparently plausible numerical result simply because a sophisticated algorithm produced it. Engineering requires intellectual participation. The student must be able to audit an answer, check dimensions and orders of magnitude, examine boundary and initial conditions, identify inappropriate assumptions, verify numerical convergence, compare results with physical intuition, and recognise what might be an AI hallucination, whether numerical, mathematical, or physical. A classroom may therefore be full of laptops, simulations, code, and AI-generated material, with students apparently working independently, and yet contain very little genuine learning. Activity is not the same as intellectual engagement. The crucial question is not how technologically advanced the classroom looks, but how much thinking the students are actually doing.

There is another misconception that deserves attention: the idea that giving students autonomy automatically produces independent learners. It does not. Autonomy without sufficient knowledge often produces frustration. Autonomy combined with poorly understood AI tools can produce dependence. And asking students to solve highly complex engineering problems before they possess the necessary conceptual foundations can turn what should be an intellectually demanding exercise into little more than trial and error. Critical judgement develops progressively. Students first need knowledge. Then they need guided practice. Then they need increasingly open problems in which they can exercise judgement. Eventually, the lecturer can step back and allow students to take greater responsibility for decisions. This is not a rejection of student-centred learning. It is a recognition that student autonomy itself has to be taught.

One of the clearest examples comes from teaching the same subject to students at different educational levels. I have taught concentrating solar thermal energy at both undergraduate and postgraduate level, and the experience makes the distinction particularly clear. At undergraduate level, the course is necessarily more foundational and exploratory. Students are introduced to the physical principles of solar concentration, optical behaviour, heat transfer, thermal losses, system components, and basic performance analysis. The objective is not simply to provide information, but to help students build a conceptual framework that allows them to understand how the system works and to explore the engineering possibilities. At postgraduate level, the same general subject can become substantially more specialised. Students already possess a stronger theoretical background, so the learning objectives can move towards research-oriented questions: optimisation, advanced modelling, sensitivity analysis, uncertainty, exergy analysis, system integration, and the evaluation of competing design configurations. The subject has not changed fundamentally. The learners have. And therefore the pedagogy should change with them. A teaching strategy that is appropriate for a first encounter with concentrating solar thermal technology may be unnecessarily restrictive for a postgraduate researcher. Conversely, exposing an undergraduate to highly specialised optimisation problems before establishing the underlying physical concepts may obscure rather than deepen understanding. This is why the question of which teaching method is best is often the wrong question. The more useful question is: best for whom, for what purpose, and at what stage of learning?

Consider a group of engineering students who are eventually expected to design a hybrid solar-wind energy system incorporating thermal energy storage. They will first need to understand solar radiation, energy conversion, heat transfer, thermodynamics, and the behaviour of the relevant working fluids. At this stage, direct instruction and guided practice are essential. Students need someone to structure the knowledge, identify the important physical relationships, and prevent misconceptions from becoming embedded. Once the foundations are secure, the supports can gradually be removed. The students can then be asked to analyse real datasets, build simulations, compare system architectures, and use AI to explore design alternatives or optimise collector configurations. They may process large meteorological datasets, investigate the effect of thermal storage size, or evaluate competing designs. But the final task should not be to ask the AI which design is best. Instead, students should have to justify their decisions using engineering evidence. They should discuss trade-offs between cost, efficiency, exergy performance, reliability, and environmental impact. They should explain why particular assumptions were made, determine whether the model is physically credible, and defend the final design when challenged. Direct instruction, student autonomy, and artificial intelligence are not mutually exclusive. They can form part of the same learning sequence.

This leads to what I consider the most important question in twenty-first-century engineering education: what conceptual, analytical, and ethical competencies should this student be able to demonstrate in order to design the energy systems of the future? Only after answering that question should we decide how to teach. The methodology should serve the engineering problem, not the other way around. Sometimes the appropriate tool will be a derivation on a whiteboard. Sometimes it will be a carefully structured laboratory. Sometimes students should work collaboratively on an open-ended design problem. Sometimes a simulation should be developed step by step. Sometimes students should use AI to analyse data or explore a large design space. And sometimes the most productive thing a lecturer can do is give a student a blank sheet of paper and ask: What would you do? That diversity is not pedagogical inconsistency. It is pedagogical judgement.

The rise of AI does not make the engineering lecturer irrelevant. It arguably makes the lecturer's judgement more important. When information is abundant and computational tools are increasingly powerful, the scarce resource is no longer access to information. It is the ability to determine which information matters, which assumptions are defensible, which results are credible, and which decisions are technically and ethically justified. The lecturer therefore becomes not simply a transmitter of knowledge, but also a designer of learning environments, a facilitator of increasingly complex intellectual work, and a guardian of disciplinary standards.

The most successful engineering education of the twenty-first century will probably not be the one that is exclusively teacher-centred or exclusively student-centred. Nor will it be the one that bans AI. And it certainly will not be the one that treats AI as an unquestionable solution to every educational problem. It will be the education that knows when students need structure, when they need freedom, when they need practice, when they need challenge, and when technology genuinely adds value. Pedagogical methods are tools. Artificial intelligence is an accelerator. But engineering judgement, and the judgement of those who teach engineers, remains the essential element. The goal is not to teach students to compete with machines. The goal is to teach them to think well when machines are in the room. And perhaps that is the most important engineering education challenge of the twenty-first century.