Engineers Who Can Finally Be Engineers – How AI Is Changing MEP Design

In MEP design offices, AI is no longer an experiment – it is a working tool. It runs complete sizing calculations, produces material take-offs, compiles technical descriptions, and takes over more and more tasks that engineers used to do, but in the role of a draughtsman or editor.

This article looks at what the profession gains from this, and where the pitfalls are, from the perspective of a practising designer. Because there are two sides to the coin: the same tool that frees up the engineer can also undermine trust in the wrong hands.

What AI is taking over

A large share of design work has always been repetitive and rule-based. These are the tasks where AI delivers the most today:

• Complete sizing calculations: heat load, pipe network and ductwork sizing, pressure drop, pump and fan selection – in minutes, with a clear, traceable derivation.
• Accurate material take-offs: itemised from the model or drawing, in a consistent format, ready for cost estimating.
• Documentation: technical descriptions, condition surveys and energy calculations compiled from project data.
• Editorial work: labelling, legends, sheet organisation, standardisation and catching inconsistencies.

These tasks used to take hours, sometimes days, out of every project. They required engineering knowledge, but not engineering judgement – the engineer was essentially working as an expensive draughtsman.

Engineers can finally be engineers

The biggest gain is not the time saved, but the fact that the freed-up time goes into real engineering work. System concepts, comparing alternatives, coordinating with the client and the construction team – in other words, responsible decisions.

This does not happen automatically, though. AI is worth as much as the engineer puts into it:

• Good prompting: describing the task precisely and specifying boundary conditions and the applicable standards is now a professional skill.
• Agent systems: multi-step workflows where each step builds on the previous one – from survey through sizing to documentation.
• Skills: the office’s own know-how, templates, drawing standards and proven solutions in a reusable form.
• Continuous development: the experience of every project feeds back into the system, making it more accurate and reliable over time.

Those who build this deliberately are not just using a tool – they are creating their own design environment. Those who only ask a chatbot now and then get a fraction of the benefits.

The final word always belongs to the human engineer

AI proposes, calculates and prepares – but it does not sign drawings and does not bear responsibility. The design is issued by the licensed engineer under their name and chamber registration number, and they are responsible for its correctness, both professionally and legally.

That is why a well-built AI workflow has human approval at every critical point:

1. Checking the input data and boundary conditions.
2. Professional review of the system choice and the main sizing results.
3. Reviewing and signing off the final documentation.

AI speeds up the work, but it does not replace engineering judgement. If a result looks “too good”, or a value does not match experience, it is the engineer who has to notice – and that responsibility cannot be delegated.

The downside: when the client checks with AI

AI has arrived not only at the designer’s desk, but at the client’s as well. It is increasingly common for a client to paste the finished design into a chatbot, ask “is this right?” – and then treat the answer as fact.

The problem is not the intention to check, but the way it is done. Typical situations:

• Missing context: the AI does not know the building’s conditions, the regulatory requirements, the budget or the history of design coordination.
• The wrong question: a leading prompt (“isn’t this boiler oversized?”) produces a biased answer.
• Confident mistakes: AI sounds just as convincing when it is wrong, so a layperson cannot tell correct statements from incorrect ones.
• Generic “good advice”: which may be true in itself, but does not apply to the project at hand.

The result is often that doubt creeps in: if “even the AI said so”, maybe the designer got it wrong. Trust in the engineer drops, even though the design is technically sound.

The engineer’s side: correcting and rebuilding trust

What takes the client a minute – pasting in the design and forwarding the answer – takes the engineer hours. Every point has to be taken seriously, since there may be a real error, so each one must be thought through, checked and answered in writing.

The correction typically looks like this:

1. Taking the AI suggestion apart: what is general truth in it, and what does not apply to this building.
2. Supplying the missing context: regulations, measured data, previous agreements, the reasoning behind the choice.
3. A clear but technically precise explanation, often backed up by calculations.
4. If needed, a meeting with the client, even in person.

The harder part, however, is trust. A single AI answer forwarded without review can damage the relationship more than its rebuttal can repair. In these situations, the engineer not only has to be right, but also has to show why – patiently, without dismissing the client’s question.

How can we work well together?

AI is neither an enemy nor a miracle cure – it is a powerful tool that creates value in responsible hands. A few simple principles help both sides:

• For clients: bring the AI’s answer to the designer as a question, not a verdict. “The AI suggested this – what do you think?” opens a dialogue, not a dispute.
• For clients: if you check with AI, give it the full context and ask open questions rather than leading ones.
• For designers: be transparent about what you use AI for, how, and where human review sits in the process.
• For designers: document the reasoning behind your decisions – it is the fastest answer to any later AI-based objection.

The designer of the future is not the one who draws the most, but the one who decides best – using AI as a colleague, not as a referee. And responsibility, as before, stays with the engineer.

---

This article was also written with the help of AI – based on the designer’s prompt, priorities and final approval. Exactly as described above.