Sam is 100% correct.
This is the artifact-outcome problem in action: assuming that because AI can produce the deliverable, it can produce the result the deliverable was supposed to generate.
In reality, the deliverable (the demand letter, the memo, the brief, the whatever) was never the value driver. It was the visible artifact of a deeper mechanism - institutional leverage, credentialed authority, relational trust, regulatory standing and/or the capacity to impose consequences - that actually produced the outcome.
You see this playing out in dozens of industries right now - medicine, finance, regulation/compliance, journalism, audit/assurance, contract negotiation, advertising/marketing, litigation - all with the same pattern: the deliverable exists within an adversarial or evaluative system where the counterparty or decision-maker isn't simply processing information; they're evaluating the source, authority, capacity consequence structure that underpins the information.
In all of those industries, the thing AI produces is a signal about the underlying value driver. This is why the "AI-as-a-disruptor" narrative continually runs into the same wall. It treats professional services businesses as if they're information production businesses.
Spoiler alert: they aren't.
Professional services businesses are trust, authority and consequence-bearing businesses that happen to produce information as a byproduct.
The brief, the letter, the deck, the whatever is a shadow on the proverbial wall. AI systems are quite good at painting shadows. But a shadow is only as powerful as the thing casting it.
Everyone is focusing on the wrong problem with AI and legal work. The problem isn't accuracy, hallucination, etc. Even if AI gets everything 100% right, it still won't matter. Here's why:
Let's say you have an insurance claim. You try to save a few bucks by NOT hiring a lawyer and instead have ChatGPT advise on strategy / write a few letters to the adjuster.
The letters are perfect. They make every argument correctly. You're already counting your settlement chickens, but then a plot twist: you get no offer, or maybe best case a low offer.
The problem isn't "bad AI" — it's that your case won't settle unless the insurance co fears losing even more money at trial. The only thing that moves the needle is the threat of getting nuked at trial.
Your letter, no matter how perfectly written, is worthless without actual litigation teeth behind it. And adjusters can tell. An unrepresented claimant sending polished letters is still an unrepresented claimant.
Insurance company logic: What's this person going to do if we don't pay? Sue us? With what lawyer? I guess they'll have to go find one eventually. Cool — we'll wait.
"Fine," you say. "I'll use AI first and hire a lawyer later if they lowball me." But now you've already shown your hand. The adjuster sized you up months ago, anchored low, and the clock's been running on your statute of limitations. Your lawyer is now playing from behind on a case the insurance company already decided not to take seriously.
The adjuster isn't scared of a well-written letter. The adjuster is scared of a trial lawyer with a filed complaint.
That's it. That's the whole game.