In this post, I consider the risks posed by medical AI, then consider how to structure contract-drafting AI.
The Risks of Medical AI
Let’s start with medical AI. If you want some gripping reading, check out this LinkedIn article, The Great Inversion: How Healthcare AI Stopped Helping Doctors and Started Replacing Them. It’s by a surgeon, John Ferguson.
It describes how AI in medicine was originally intended to help doctors with their crushing administrative burden. Instead, “AI handles the clinical reasoning so the physician can focus on supervising the AI.” Here’s how Ferguson describes that:
The “loop” the human occupies isn’t clinical reasoning anymore. It’s quality control on someone else’s work. And that someone else can’t smell ketoacidosis, can’t see the tremor in a patient’s hands, can’t hear the catch in someone’s voice when they say “I’m fine” but they’re not fine. That someone else doesn’t carry malpractice insurance.
Then consider how the diagnostic role changes the doctor’s work. Ferguson again:
You’re scanning for errors in reasoning you didn’t generate. The vigilance required to catch AI errors is higher than the effort required to just do the reasoning yourself, but it feels easier because you’re reading instead of generating.
For doctors, that poses the risk of “clinical deskilling.”
Ferguson then considers the goal of AI being given a broader role:
An AI system that does your charting saves you time. An AI system that does your thinking saves the health system your salary.
Now ask: who decides which AI a health system purchases? Who evaluates, selects, and implements clinical AI tools?
Not the physician whose judgment is being replaced. The C-suite. The administrators. The people optimizing for cost per encounter, throughput, and margin.
What Might Contract-Drafting AI Be?
So, does contract-drafting AI poses the same risks? Well, yes and no.
Let’s start by considering the context. First off, what do I mean by contract-drafting AI? I don’t know. AI is currently being applied to contract review. As described in this LinkedIn post by Scott Stevenson, CEO of Spellbook, contract-review AI is also being used by those using precedent contracts as the starting point for drafting a contract for a new transaction.
I’m not much interested in contract-review AI—it involves responding to the world as it is. By contrast, contract-drafting AI has the potential to allow us to create a better, more efficient world. So I have something more ambitious in mind for contract-drafting AI, something that allows the user to make a fresh start.
The Obstacles to Contract-Drafting AI Are Modest
Contract drafting is very amenable to applying AI. The obstacles are less imposing than they are in medicine. For one thing, the universe of contracts is way less complicated than medicine. Contracts are just documents!
And one important constituency is missing from contracts: patients. “The world of commerce” is too diffuse a constituency for anyone to take seriously. That ought to make it easier to give AI a significant role.
Lawyers Are Willing to Sideline Themselves
With that established, what are the risks that contract-drafting AI would reduce the role of the lawyer from setting parameters for a new contract to vetting parameters set by AI?
That risk is very real. Because of generations of copy-and-pasting, on faith, from precedent contracts and templates of questionable quality and relevance, deal lawyers are prone to having accrued significant cognitive debt. (This 2025 MIT study defines cognitive debt as “a condition in which repeated reliance on external systems like LLMs replaces the effortful cognitive processes required for independent thinking.”) I’ve encountered lawyers who balk at being presented with a broad decision tree for a new transaction. Instead, they prefer being given the attenuated decision tree that comes with copy-and-pasting—the decision tree for some other deal.
So lawyers would probably be complicit in their own sidelining.
Garbage-In, Garbage-Out
We also have to consider the dysfunction of mainstream contract drafting. I’ve devoted my career to chronicling it. If the dysfunction made its way into contract-drafting AI—which it would, without strong controls—the result would be garbage-in, garbage-out. It would be like basing diagnostic medical AI on the principles of Hippocrates and Galen.
Balancing the Risks
Because contract drafting AI presents risks different from those presented by medical AI, I contemplate contract-drafting AI that in one respect invites lawyers to do better but in another respect constrains them.
My fantasy contract-drafting AI would present the user with a broad decision tree for a given transaction and offer them whatever guidance they might need. You can’t be bothered to think through what you want a new transaction to look like? Then go do something else.
But my contract-drafting AI wouldn’t indulge the user in copy-and-pasting the dysfunction of mainstream contract drafting. Instead, after determining what decision tree applies, the user would be speedily presented with text reflecting those choices. What they’re given would be clear, concise, and relevant.
What, you think my contract-drafting AI looks very much like contract automation? (Contract automation is what I’ve been doing at Adams Contracts.) If all you have is the hammer that is AI, everything looks like a nail, but different problems require different solutions. I’m confident it would be a straightforward matter to come up with a hybrid AI–contract-automation solution.
Making Contracts Better
How AI is applied to contracts depends largely on what motivates those calling the shots. If it’s the C-suite, the administrators (to use Ferguson’s terms), the path to quickest returns would probably be to turbocharge the current contracts process, focus on lopping heads, and limit those that remain to checking the AI’s output.
I hope that instead, we take the opportunity to make contracts better.
