Author’s note: As you may have noticed, I took a short Summer hiatus from my articles. I hope you have also enjoyed some time to refresh. Well, I’m back and I hope you enjoy my new article, where I attempt to pierce the efficiency narrative and explore the value that we abandon in greater pursuit of efficiency.
A hotel decides it no longer needs its doorman. The doorman opens the door. An automatic door opens it too, costs less, and never calls in sick. A spreadsheet calculates the savings, and one morning there is a sliding glass panel where a person used to stand.
The spreadsheet nor the C-suite could not see the rest of the job. The doorman hailed cabs, carried bags, and remembered the regulars by name with a cheery grin. He watched the lobby and turned away people who had no business being there. He made arrival feel like an occasion. None of that was in his job title, so none of it survived the analysis. Rory Sutherland, the ad man who thinks like a behavioral economist, calls this the Doorman Fallacy. You define a job by its narrowest describable task, you automate that task, you book the savings, and you never notice what you destroyed, because the many parts of its sum, some of intangible value, never made it into the bottom line.
We are about to make this mistake at a scale never imagined, because we have built machines that are very good at reproducing the describable task. And, to make it worse, there’s yet another psychological trap that makes the fallacy almost impossible to catch in time.
If this sounds interesting to you, read on…
This substack, LawDroid Manifesto, is here to keep you in the loop about the intersection of AI and the law. Please share this article with your friends and colleagues and remember to tell me what you think in the comments below.
The Doorman Fallacy Reveals the Value of the Intangible
The Doorman Fallacy lives in the gap between what a job is called and what a job is. The job title covers the visible, obvious, automatable part. But, the value sits somewhere harder to measure.
Now, let’s apply this to law firms. A client looks at a lawyer and sees document production. A contract materializes. A motion is drafted. An answer is filed. These are the describable tasks or skills, and they are part a capable model now does in seconds. AI drafting a document is the automatic door.
What the analysis can’t price in is the rest. The judgment that picks which problem to solve before a word gets drafted. The clause left out on purpose, because including it would start a fight the client can’t afford. The call that talks a founder out of the lawsuit she walked in demanding, because the lawyer sees the counterclaim coming and she doesn’t. The risk the lawyer absorbs by putting her name on the work and her license behind the advice. The reassurance, late at night, that the thing keeping a client awake is survivable. The decision about what not to do, and how to do it when, which is what experience and judgment cultivates.
But, lawyers solve problems, they aren’t task rabbits. The document is the artifact of the solution. It is not the solution itself. The contract is the doorman’s door: the most visible thing, and the least valuable of it. When we let the profession be described as task completion or step-by-step skill description, we hand AI its easiest win, because knowledge work-oriented task completion is the game the machine was built to win.
The Amnesia Effect Blinds Us to AI’s Limits
Michael Crichton noticed something about reading the news. If a reporter gets something wrong in a field you know cold — not a typo, but gets the cause reversed or the basic facts misstated, you shake your head in disappointment. But, if the error pops up in a story about foreign policy or medicine or some science beyond your expertise, you read it as if it were true. You forget, in clicking to the next news story, that the same news channel had just failed you. Crichton called this the Gell-Mann Amnesia effect, named half in tribute to a physicist friend who liked the idea.
Some might recognize the Dunning-Kruger effect here. It is adjacent but distinct. Whereas with Dunning-Kruger a person without expertise in a subject is led to be overconfident about their grasp of it due to their ignorance of its complexity, with Gell-Mann an individual affords a source credibility on subjects they are unfamiliar with even when the same source has demonstrated its unreliability in a subject they are knowledgeable about.
Now, think about how this amnesia effect applies to AI. A lawyer asks a model a question about law and catches the invented case, the rule stated with full confidence and quoted slightly wrong, the citation to an opinion that holds the opposite. She winces, the way experts wince, and resolves to be more vigilant. Then she asks the same model to explain a failed bridge’s engineering, summarize medical records about a client’s treatment for multiple myeloma, or architect a variation on a transformer model, and she reads the answers as if they were true. The errors may be there. They just became invisible the moment the subject left the field where she could see them.
The machine earns your trust exactly where you can’t audit it. Its fluency holds steady across every subject. Your ability to check that fluency falls away the moment you step outside your own are of expertise. The model’s confidence never drops, so the only thing that flags an error is your own knowledge, and knowledge is what you lack in every field that isn’t yours.
How These Two Biases Reinforce Each Other
Take a client deciding whether to keep paying a lawyer. Through the Doorman Fallacy, the lawyer looks like an expensive scrivener, surely something the new tools can handle better now. Through the Gell-Mann Amnesia effect, the AI’s legal output looks fine, because the client can’t see the errors a lawyer would catch. Although a client who is a physicist recoils when an AI model doesn’t properly check for dimensional consistency, the same scientist may believe that a free or cheap AI-generated estate plan is fine.
The fallacy says the lawyer is replaceable. The amnesia effect says the replacement is good enough.
The thing that would break the spell is the same in both cases. Only an experienced eye can see what the doorman really does, and only an expert eye can see where the machine really fails. The value you can’t see and the error you can’t see are one blind spot. And that expert eye is the thing being shown the door in the name of efficiency.
Closing Thoughts
The efficiency narrative is compelling. Efficiency is always measured against a defined, measurable task. It optimizes what’s written down and stays blind to what isn’t. A better metric or a better benchmark (we’ve seen a proliferation of those lately) won’t fix it. To measure something you first have to define it, and defining it narrows it to the visible, objective part, the part the doorman shares with the sliding door. This is the danger of slogans like, “What gets measured gets managed,” often attributed to management consultant Peter Drucker; it encourages the well-intentioned to race to measure everything and forget the rest. But, the unmeasured and immeasurable value and the unchecked error are the two things efficiency can’t see by design.
So the drive for efficiency doesn’t miss reality at the edges. It misses it at the exact spots where inherent value and the risk of its removal overlap. The value the Doorman Fallacy throws away and the error the Gell-Mann Amnesia effect waves off are the same kind of thing: real, costly, and invisible to anyone who doesn’t know better.
For all of us, that points to an economy reorganizing around distilling the professions and industries to their nominal tasks. Automate the task. Trust the automation most in the places where no one in the room can check it. The visible savings speaks for itself. The invisible value drains out. The errors collect in blind spots and compound, the way unseen things do, slowly and then all at once — as we’ve already seen with multiple examples of AI agent deception and security breaches. By the time the lobby feels less safe and welcoming, no line item will tell you why, because the reason was never a line item to measure and streamline.
None of this is a per se argument against AI, and none of it is an argument against efficiency. Both are worthwhile. AI on the measurable, definable task is a gift, and I build with it every day.
My argument is about not letting the AI-ification of tasks, skills, and workflows be the end of the story, and for keeping the expert eye and our humanity in the narrative as well.
We need the person who can see what the doorman does and where the machine fails, who can tell a draft that’s quick and fluent-looking from a draft that one can stand behind, and who is worth paying for value that will never show up cleanly on a spreadsheet.
Tom Martin is CEO & Founder of LawDroid, Adjunct Professor at Suffolk University Law School, and Author of AI with Purpose: A Strategic Blueprint for Legal Transformation (Globe Law and Business). He is “The AI Law Professor” and writes his eponymous column for the Thomson Reuters Institute.
By the way, if you would like to meet other legal innovators in person, and enjoy an exceptional awards gala celebration, don’t miss the Oscars of Legal Innovation: the American Legal Technology Awards, this October 25, 2026, in Boston. Reserve your Early Bird tickets today and save $100.
