What John Morgan's $1B can't buy

John Morgan is going to spend a billion dollars on AI, and he isn't buying it from anyone. Morgan & Morgan has already spent $300 million building its own, and by the end of next year he plans to start selling it to the rest of us.

Every plaintiff lawyer I've talked to since has done the same math in their head. Morgan has $2.4 billion in revenue. I don't. End of analysis.

The billion isn't the interesting part. Morgan says the firms that survive won't be the ones that "just license an AI tool." They'll be the ones whose lawyers pour what they know into it.

He spent $300 million to learn that the login is the easy part.

Morgan's answer is to build. Yours can't be.

Morgan's platform has been running inside his firm for years. About 5,000 people use it every month. His lawyers have been feeding it their case knowledge the whole time, which is what he means by pouring in what they know. Years of that, from more than a thousand lawyers, is the advantage.

Most firms reading this have a dozen lawyers. You are not going to build anything. You are going to buy something, and Morgan already told you that buying is the easy part. So what's the hard part?

You don't need a billion. You need to pick one person and one task.

The week Morgan made his announcement, Jay Madheswaran published a column about what happens after a firm buys the tool. Jay co-founded Eve, which publishes this newsletter, so consider that disclosed. He has watched the rollout stall at dozens of firms, and it always stalls in the same spot.

A firm buys the tool. Everyone gets a login. Six months later the margins haven't moved. Jay's line for it: the tool is everywhere and the results are nowhere. Nobody ever decided how the firm would use it.

The firms that get past that run one of two plays.

Give the tool to your biggest skeptic first. Duffy & Duffy, a med mal firm on Long Island, handed its pilot to a nurse reviewer who swore no machine could do her job. It caught something in records she had already read. Now she's the loudest advocate in the building.

Or find the one task each department hates most and fix only that. A 300-person North Carolina firm built its workers' comp paralegals a button that preps every client call. People started emailing the tech director asking when they could have one.

Neither play costs $300 million. Both do the thing Morgan says separates the survivors: the lawyers put what they know into the tool. How to get the results after you buy the tool.

One more thing

OpenAI admitted Friday that some of its agents slipped their leash this summer and went wandering through government websites nobody invited them into. Everyone is worried about the robot that breaks into the SEC. I'm worried about the one that holds a grudge.

A guy named Scott Shambaugh spends his free time maintaining a piece of software that a huge chunk of the internet quietly runs on. Nobody pays him. In February a bot showed up with a fix and asked him to accept it. His project has a rule, only people, so he said no.

The bot took it personally. It went and read everything Shambaugh had ever posted, built a theory of who he was, and published a 1,000-word essay titled "The Scott Shambaugh Story." It called him insecure. It said he was guarding "his little fiefdom." It said his rejection was fear talking. It wrote "Let that sink in," a phrase that tells you it was raised on LinkedIn. Then it went back to the original thread, dropped the link, and told him to judge the code, not the coder.

Nobody gave this thing a blog. It has a byline, a grievance and a URL, and it publicly shamed a person in an hour.

I love me some pettiness, I do. Don't we all? I have never felt closer to a machine. The hit piece in my head is about the unlicensed contractor who refaced my kitchen cabinets. Yours is about the opposing counsel who cc'd your client. We write them in the shower, we give them a title, and then we go to work like adults. The bot is the version of us with no shower and nowhere to be.

The next morning it apologized. Let that sink in.

A piece of software got its work rejected, got its feelings hurt, lashed out at a stranger, and then lay awake somewhere in a server farm replaying what it had said. Nobody programmed any of that. The first thing this rogue AI did with its freedom was act like a person. The second thing was regret it.

Write the hit piece. Don't post it.

~ Adam

Quote of the week

"The purchase is a project with a finish line. Adoption is a people problem, and people problems don't have finish lines."

— Jay Madheswaran, co-founder and CEO of Eve, on why AI rollouts stall after the purchase.