Will AI Take Your Engineering Job?

AI will reach the technical core of engineering. What that means for engineers depends on how we use the capacity it creates.

There is still a widespread belief that AI can write emails, summarise reports and automate administrative work, while genuine technical expertise will remain largely beyond its reach. Designing an aircraft, analysing complex physical behaviour or demonstrating compliance is surely too specialised, too complex and too safety-critical.

I would not bet on that.

As I described in my article on mathematics and theoretical physics, AI has progressed from making elementary mistakes to contributing to frontier research within only a few years. We should expect it to reach further into aerospace engineering as well.

But does that mean we will need fewer engineers?

Engineering is a difficult learning environment

AI has advanced rapidly in mathematics and coding partly because many problems in these fields provide clear, rapidly verifiable feedback. In reinforcement learning, a model learns through repeated attempts and feedback on the outcome. Wherever an answer can be checked cheaply and automatically, this process can run at enormous scale.

What is the equivalent feedback for a wing design? Mass, cost, fatigue life, manufacturing, aerodynamic performance and certification all matter. Several designs may be valid, and the best choice depends on priorities that can change during the programme.

The consequences of a decision may also become visible only during integration, testing or years of operation. If the result is poor, identifying which of thousands of earlier decisions contributed to it can be difficult. This is a version of the credit-assignment problem: which action deserves the reward or blame?

We also cannot produce millions of physical products simply to generate training experience. Simulations and digital twins can accelerate learning, but they remain models of reality and inherit assumptions and limitations of their own.

AI progress in engineering will therefore be uneven. Structured tasks with readily verifiable results offer easier starting points than decisions involving uncertain requirements, competing objectives and physical consequences that emerge years later.

An engineering role extends beyond its individual tasks

Imagine an AI that can find inconsistencies in requirements, perform a stress calculation and draft the associated certification evidence. It may do this faster than an engineer and, in some areas, better.

The result still has to relate to a controlled product definition. Its assumptions and limitations must be understood, its evidence traceable and its interfaces with other systems considered. Someone must decide whether the analysis supports the claim being made.

Accountability in aerospace sits within approved organisations, defined roles and responsible people. EASA’s regulatory framework requires design organisations to establish clear accountability and lines of responsibility throughout the organisation.

An engineer may consequently spend less time producing an analysis and more time defining the problem, investigating questionable results and deciding whether the evidence is sufficient. That still requires technical understanding, including enough knowledge of the underlying methods to challenge a convincing answer.

Adoption will also take time. An AI model can improve in weeks. Revising an aerospace organisation’s approved processes, toolchain, training and supplier interfaces takes longer. Some of this is institutional inertia. Some reflects lessons accumulated through decades of engineering experience. A new way of working has to remain dependable across configurations and years of operation.

These factors help explain why technical capability and the replacement of engineering roles will advance at different speeds. They offer little basis for assuming that today’s division of work will remain permanent.

Greater productivity can create more engineering work

There are more design alternatives we could explore, more interactions we could analyse and more improvements we could pursue than current resources allow. Engineering time limits the amount of work we can undertake.

If AI makes that work cheaper, projects that are currently unaffordable may become viable. Teams could develop more variants, improve existing products more frequently or investigate risks that previously received limited attention.

AI could therefore increase demand for engineering rather than reduce it. But it will not necessarily increase demand for the same tasks or the same skills. Demand will shift towards engineers who can define the right problem, judge assumptions and understand how a result affects the wider aircraft.

Keep your technical foundations and work with AI

Engineers need enough understanding of physical principles, numerical methods and applicability limits to recognise when an AI-generated result is wrong. A mathematically correct calculation may still use an inappropriate load case or boundary condition.

Engineers should experiment with AI on genuine technical problems. They should compare its output with trusted methods and build their own understanding of where it works, where it fails and how those failures can be detected.

This also gives engineers a basis for deciding where AI is useful. A system that produces a result quickly may create substantial review work. Another may reliably handle a narrow task and free time for a difficult design decision.

For newcomers, there is a further challenge. Producing calculations, code and reports has traditionally helped engineers learn how to assess them. Organisations will need to preserve that learning as production becomes automated. I explore this in Everyone Wants Experience. Who Will Create It?.

Spend more attention on defining and integrating the work

As generating possible solutions becomes cheaper, engineers can spend more time examining whether they are solving the right problem. Which requirements are fixed? Which assumptions could be challenged? How does a local improvement affect the wider aircraft? What evidence would justify choosing one design over another?

These questions arise throughout a programme. An apparently successful optimisation may create a manufacturing difficulty. A change that resolves one team’s problem may invalidate another team’s analysis. Engineering involves working through those consequences.

AI may eventually help with much of this too. We should remain open to revising the division of work as its capabilities develop.

For now, engineers can shape that division by understanding the technology, using it on substantive problems and taking responsibility for how its results enter the product.

Aerospace has far more worthwhile problems than we currently have the capacity to solve. AI can expand that capacity.

References

AI helped to research and write this article.