Mechanical engineer — tasks, one by one

Requirements and concept

Requirements come from customers, standards and the conditions a part will face, and the trade-offs are judgements someone answers for. In NASA's use of generative design, the engineer starts from the mission's requirements and defines where the part connects before the software proposes anything. US projections expect mechanical engineering employment to grow 11 percent from 2025 to 2035, much faster than average.

A projection counts jobs, not how much of the concept work tools now do; the NASA account is one organisation's use.

CAD modelling and generative design

Software now proposes the geometry once the engineer has defined the constraints: NASA reports generative design producing complex structures in as little as an hour or two, and parts up to two-thirds lighter, while the engineer who pioneered it says the algorithms need a human eye because they can make structures too thin. Models can also write CAD scripts that compile, though a compiled script is not a part that meets its specification.

NASA's figures come from the engineer who promotes the method; the benchmark measures whether scripts run, not whether parts are right.

Simulation and analysis

Models can drive simulation software but rarely reach a correct answer on their own. On FEABench, the best strategy produced executable calls to a commercial simulation tool 88% of the time, yet the models computed a valid target within 10% of the right answer for almost none of the problems. Setting up the model is being assisted; judging whether the number is right stays with the engineer.

A benchmark by a model developer on a small set of problems; it does not measure engineering work on real products.

Prototypes, testing and failure investigation

Testing and diagnosis happen on physical parts, and the analysis behind them depends on reading drawings and rules. On DesignQA, a benchmark built on real Formula SAE rules and CAD, multimodal models struggled to retrieve the relevant rules reliably, to recognise components in CAD images and to analyse engineering drawings. In NASA's account, generated parts are still analysed with NASA-standard validation software and processes before use.

The benchmark tests models from 2024 on one competition's rules, with authors including a design-software company; it does not measure testing work.

Safety, compliance and sign-off

Rules keep this work with people and add to it. From 14 January 2027 the EU requires safety components and machinery with machine-learning, self-evolving behaviour ensuring safety functions to go through a third-party conformity assessment, and risk assessment must cover how that behaviour is intended to evolve. In Texas, plans and specifications for projects built or used in the state must carry the license holder's seal.

The EU rule governs how machines are assessed, not whether engineers' tasks are automated; in Texas, engineers who design a company's own manufactured products are exempt from licensing.

Manufacturing and equipment integration

Automation adds work here. The US statistics bureau expects mechanical engineers to be needed to help integrate more complex automation machinery into existing systems as manufacturing processes incorporate it.

This is a projection's stated reasoning for one country, not a measurement of how much integration work engineers now do.