When the Lawyer Calls, Your AI Better Have Receipts
Graph RAG is usually sold as better search. In engineering, it may become the defense file that explains why a decision was made years after the project is done.
Your AI better be able to explain itself in a deposition.
Not in a demo.
Not in a sales deck.
In a deposition.
The demo is easy.
The subpoena is where the system gets tested.
AEC is a long-tail liability business.
Something you stamped in 2024 can become a lawsuit in 2031.
Seven years later, your lawyer is not going to ask:
“Did the AI seem smart?”
They are going to ask:
“Why did you specify it this way?”
And the answer better not be:
“I think someone on the team remembered seeing that in an email.”
Today, most engineering firms defend decisions by digging through old PDFs, emails, RFIs, meeting notes, specs, code comments, and half-remembered conversations.
Then a grey-haired principal tries to reconstruct the logic under oath.
Terrible system.
Graph RAG changes the game for engineering firms, but not for the reason most people think.
Everyone talks about better search.
Fine.
Search is useful.
But the real value is the evidence trail.
When the system says:
“This detail applies to this project.”
It should also be able to show:
The spec section.
The code reference.
The AHJ ruling.
The RFI response.
The submittal comment.
The prior project precedent.
The internal standard.
The person who approved the exception.
Not vibes.
Not “the model said so.”
A defensible chain of reasoning.
That matters in MEP.
We do not just need AI that can answer questions.
We need AI that can prove why the answer was allowed.
Why this equipment was selected.
Why this clearance was acceptable.
Why this load assumption was used.
Why this exception was made.
Why this note ended up on the drawing.
The firms that figure this out will not just move faster.
They will reduce risk.
They will train younger engineers better.
They will stop losing institutional knowledge every time someone retires.
They will stop treating project memory like a scavenger hunt.
Graph RAG is usually sold as a knowledge management tool.
In engineering, it is also a litigation defense file.
Build the system so it can answer the lawyer seven years from now.
Not just the PM this afternoon.
Disclaimer: I’m not a lawyer. Please do not treat this as legal advice.
I’m using litigation risk as the thought experiment.
The point is simple:
If AI is going to influence engineering decisions, the system better preserve the evidence trail behind those decisions.