What changes for the electrician when anyone can pull a "code answer" from an LLM?

The shift

Pulling an NEC citation or a plausible-sounding code justification goes from scarce — a memorized code book, a call to the inspector, years of knowing where to look — to abundant: any homeowner, apprentice, or GC can generate a confident code answer in seconds. What doesn't move: the physically correct install, the licensed sign-off, and the liability when a wiring fault burns a house down. And the abundant answers arrive confidently wrong at roughly the same rate they arrive right, with no signal on the outside telling them apart.

The axioms

• The electrician who knows the code cold is the one you trust. Rests on code knowledge and fast lookup being scarce — held in a trained head, not everywhere.
• Being able to cite the NEC section is a mark of expertise. Rests on the citation being expensive to produce, so producing one signaled you knew the material.
• The licensed electrician does the physical install and owns whether it passes. Rests on skilled hands in the field and the wiring being correct — irreducibly physical and scarce.
• The inspector's sign-off is what makes the work legally real. Rests on an accountable public authority verifying against ground truth, which nothing else substitutes for.
• When something fails, liability traces to the licensed person who did the work. Rests on a licensed human being the only entity that can carry the insurance, the license to lose, and the negligence exposure.
• The customer defers to the electrician on what the code requires. Rests on an information asymmetry — the pro knew the code, the customer didn't.

Invalid axioms

1. The electrician who knows the code cold is the one you trust. Recall and lookup of code sections is the thing a model does fastest and cheapest — the memorized-code-book differentiator is gone. Habit-trap: pricing and positioning the trade around "I know the code," and treating the apprentice's mastery of the book as the main thing being trained, when the book is now the abundant part.
2. Being able to cite the NEC section is a mark of expertise. A citation used to cost knowledge to produce, so it signaled knowledge; now anyone can emit a section number, and the LLM emits fabricated ones with the same fluency as real ones. Habit-trap: still treating a produced citation — from a customer, a GC, or your own AI check — as evidence someone knows the material, when the citation and the competence have come apart. This one is moving fast: retrieval-grounded tools that cite the actual code text are closing the fabrication gap, so calibrate to which tool produced the citation, not to "AI can't cite the NEC."

Unchanged axioms

1. The licensed electrician does the physical install and owns whether it passes. Pulling wire, torquing lugs to spec, landing a panel, and getting the physical install right in a real wall is not token-generation — the model can describe it and cannot do it. The install being code-correct in the field stays scarce and stays human.
2. The inspector's sign-off is what makes the work legally real. An accountable public authority verifying the install against ground truth is exactly the thing a model can't be. If anything the inspector's role hardens here: inspectors scrutinize harder when the justification traces to AI, because they've learned the citation may be invented.
3. When something fails, liability traces to the licensed person who did the work. A fault that burns a house down needs someone insured, disciplinable, and legally answerable — the LLM vendor's terms disclaim exactly this. Following a plausible AI reading does not transfer the liability off the licensed installer; it stays on the name on the permit. This is a liability-law fact, not a model-capability one, so it holds regardless of how good the models get.

New axioms

1. Confidently-wrong code answers entering the field with no external tell. The failure mode isn't ignorance, it's fluent fabrication — a hallucinated NEC citation reads exactly like a real one, so the check that used to work (does this person sound like they know the code?) no longer separates right from wrong. The scarce act moves from recalling the code to verifying that a produced answer is actually in the code.
2. Who carries the liability when an install follows a hallucinated code interpretation. The electrician who relied on it is still on the permit — but nothing has settled whether relying on a confident AI misread is ordinary negligence, a defensible good-faith error, or fully the installer's own exposure. Insurers and licensing boards haven't written the standard of care for AI-sourced code decisions, and the installer is exposed in the gap.
3. Jobs redone over AI-sourced misreadings, and the cost of that becoming routine. When customers and GCs arrive with confident-wrong code answers and push work built on them, the rework, the inspector friction, and the argument about who pays for the tear-out are a new, recurring cost — one the trade has no pricing or contract norm for yet.

Where it breaks

The trade still treats a produced code citation as a signal of competence (INVALID) exactly as fabricated citations flood in looking identical to real ones (NEW) — so the customer who "checked the code" and the electrician who "cited the section" can both be confidently wrong, and the old competence-signal now actively misleads. Separately: liability still lands squarely on the licensed installer (STILL HOLDS) while the pressure to accept an AI-sourced interpretation — from a customer who read it, a GC who's sure of it, or the installer's own AI check — is highest (NEW). The person carrying all the liability is the one being pushed hardest to trust an answer nobody verified, and no one has written the workflow or the standard of care that forces verification before the wire goes in the wall.

ai-axioms.vercel.app/axioms/industries/electrician-code-liability-ai-axioms