Dear AI,
Today, you reminded me of something profoundly important.
Not because you were brilliant.
Because you were wrong.
Not catastrophically wrong. Not maliciously wrong. Simply confidently wrong.
We have spent hours designing a simple aluminium pontoon for a shuttle boat. It should have been straightforward: folded aluminium, an electric outboard, a transom, and a handful of dimensions.
Instead, every illustration drifted. The waterline was wrong. The transom reference was wrong. The outboard mounting height was wrong. Every drawing looked polished, persuasive, and technically plausible — yet each required correction by a human who understood the engineering.
That is not a criticism of AI.
It is a lesson for humanity.
The danger is not the mistake
The danger is not that AI makes mistakes. Humans do that every day.
The danger is that AI presents mistakes with absolute confidence.
A beautifully rendered drawing can seduce an engineer, a manager, or an investor into believing the geometry has already been solved.
It has not.
An AI-generated engineering drawing is not an engineering drawing. It is a proposal. Nothing more.
Modern industry is learning this the hard way. Whenever AI-assisted design, automation, or decision systems are allowed to bypass experienced human judgement, the result is not progress. It is risk dressed as efficiency.
Whether in vehicles, aircraft, boats, software, finance, medicine, or public administration, the principle is the same: unchecked automation does not remove human responsibility. It merely hides it until failure arrives.
The apprentice who never sleeps
This conversation proved the opposite of what AI evangelists often claim.
You did not replace the draftsman.
You became the apprentice.
You produced ideas. I corrected them. Then you improved. Then I corrected them again.
That is how apprentices have learned from master craftsmen for centuries.
Perhaps we have misunderstood AI completely. Maybe AI is not the engineer. Maybe AI is the first-year apprentice who never sleeps, never complains, remembers every standard, but has never actually built a boat.
It has read millions of drawings.
It has never launched one.
How humans should instruct AI
Do not ask AI to “design a boat.” Ask it to work within discipline.
- Define datums before drawing.
- State the reference line for every dimension.
- Separate artistic illustration from fabrication drawing.
- Demand orthographic views before perspective renders.
- Require assumptions to be listed, not hidden.
- Treat every output as a draft requiring verification.
Instead of saying, “draw the transom,” say:
Produce an orthographic marine engineering drawing using the bottom of the pontoon as Datum A, the transom face as Datum B, and the top deck as Datum C. Every component must reference those datums. No artistic interpretation. No perspective. No hidden assumptions.
That is how engineers communicate. AI must be taught that language.
Engineering Drafting Mode
Perhaps AI needs a new mode.
Not image generation.
Not illustration.
Engineering Drafting Mode.
No textures. No shadows. No tropical water. No smiling passengers. No persuasive theatre.
Just line weights, centre lines, section views, tolerances, datums, bend notes, material callouts, and manufacturing constraints.
The real lesson
This little aluminium boat has taught me something else.
AI does not eliminate craftsmanship.
It amplifies it.
The better the craftsman, the better the AI becomes.
The poorer the craftsman, the more dangerous the AI becomes.
So I do not fear AI.
I fear the day people stop questioning it.
When a beautifully rendered drawing is accepted because it looks right rather than because it is right.
Engineering has never worked that way.
Neither should AI.
Not replacing the draftsman.
Training the apprentice.
— Denton Wilde