The tasks AI absorbs first are the tasks junior engineers learned judgment on. Redlines, calculation checks, boilerplate drafting, submittal logs, quantity takeoffs. The repetitive work that looked like low-value labor was also the apprenticeship. Automate it without replacing it and you will get faster deliverables now and a shortage of engineers who can evaluate them in eight years.
This is the AI problem in our industry that gets the least serious attention, because its cost arrives on a delay long enough that nobody currently making the decision will be accountable for it.
What the tedious work was actually teaching
Nobody designed the traditional path as a curriculum, which is why it is easy to dismantle without noticing.
A first-year engineer who checks two hundred calculations develops a sense for when a number is wrong before they can articulate why. One who marks up redlines for a year absorbs a senior engineer's judgment by seeing which choices got corrected and which did not. One who assembles submittal logs learns how a project is actually put together, in a way no orientation session conveys.
That is calibration, and it is built by repetition against feedback. It is the difference between an engineer who can produce a report and one who can look at a report and say something is off here.
Now consider what happens when the tool does the first pass. The junior engineer reviews AI output instead of producing work. Reviewing is a genuine skill, but it is a different one, and it is much harder to learn first. Reviewing without calibration is not reviewing. It is approving. The engineer has no basis for suspicion because they have never developed the pattern library that generates suspicion.
The specific risk, stated plainly
A first-year engineer with a capable AI assistant produces work at roughly the level of a third-year engineer, and it looks good. Leadership observes accelerated capability, which is real, and reasonably concludes the tool is working.
What is not observable in the same timeframe: that engineer's ability to detect a wrong answer has not developed at the same rate as their ability to produce a plausible one. The gap is invisible while the work is being reviewed by someone senior. It becomes visible in year six or eight, when they are the reviewer, and the firm discovers it has a cohort that can generate competent deliverables and cannot reliably tell when one is wrong.
That is a licensure-relevant problem, not just a staffing one. Responsible charge assumes a licensee capable of independent professional judgment. A generation trained primarily to edit machine output has had less practice forming that judgment, and the standard of care does not adjust for how someone was trained.
What I would not do
Withhold the tools from junior staff. Tempting and wrong on three counts. It is unenforceable. They will use them privately, unsupervised, which is the worst arrangement available. It puts them at a disadvantage against peers at other firms. And it makes your firm visibly unattractive to exactly the people you are trying to recruit.
Preserve the tedious work as a rite of passage. Equally wrong. Making people do work a machine could do, purely for character formation, is both insulting and uncompetitive. Your clients are not paying for apprenticeship overhead, and your competitors are not carrying it.
The task is not to protect the old path. It is to replace what the old path delivered, deliberately, at lower time cost.
What actually works
Make them produce before they review
The single most effective rule, and it costs almost nothing: early-career engineers may use AI to draft and to check their understanding, and may not submit anything they could not have produced and defended themselves.
This is not about output. It is a constraint that forces the underlying competence to develop, and it is enforceable through a simple habit, the reviewer asks the junior engineer to explain a choice in the document. An engineer who cannot explain why a recommendation is what it is has learned something about their own preparation, and so have you.
Teach calibration explicitly, since it is no longer incidental
The judgment that used to accumulate as a side effect now has to be taught on purpose. The good news is that it can be taught faster and better than repetition taught it.
Run error-hunting sessions on real firm documents seeded with plausible mistakes. A bearing capacity that is reasonable but wrong for the site, a flattened project-specific exception, a fabricated standard citation. Ask junior staff to find them. This builds the pattern library directly, in hours rather than years, and it works: people who have caught a fabricated citation once never trust one again. The catalogue to teach from is in what AI gets wrong in construction documents.
Have senior engineers narrate their reasoning
Redlines used to transmit judgment silently, the junior engineer inferred the rule from the correction. With less redlining happening, that channel narrows, and it has to be replaced with something explicit.
The cheapest substitute is a senior engineer saying out loud why they changed something. Fifteen minutes on a marked-up draft, explaining the reasoning rather than just the correction, transfers more than the redlines did, because it states the rule instead of leaving it to be inferred. It requires senior staff to have fifteen minutes, which is a scheduling decision leadership controls.
Give them the hours the tool freed
Here is where the payoff is available and mostly unclaimed. If AI removes twelve hours of formatting from a junior engineer's month, those hours can go to site visits, sitting in client meetings, shadowing a senior engineer through a difficult judgment call, or attending the design review they used to be too busy to join.
Those experiences build judgment far more efficiently than checking calculations ever did. The traditional path was not good at developing engineers; it was just what fit around the work that had to get done. Removing that work is an opportunity, provided the recovered time is deliberately reinvested rather than absorbed into higher utilization targets, which is a specific instance of deciding what happens to recovered hours.
The tedious work was never the point of the tedious work. Judgment was. If you take the first one away, you still owe your people the second.
What to say to them directly
They are already wondering whether this profession still needs them, and the silence is worse than the conversation. Say what is true: the entry-level task mix is changing, the bar for what a second-year engineer is expected to handle is rising, and the firm still needs engineers who can think, which is why you are investing in their judgment specifically rather than just handing them a login.
Then make that investment visible. A firm that can describe how it develops engineers in this environment has a recruiting advantage over one that cannot, and the people you most want to hire are asking.
If you are working through what this means for your development program and your review process, that is exactly what our team enablement engagement addresses, built on your own projects. Start a conversation.
← Back to insights