As Legalweek kicked off its furious four-day legal tech bonanza, legal AI darling Harvey festooned the annual awards gathering with black balloons emblazoned with the Harvey logo. I didn’t go to the awards, but I heard that attendees were asked to pop these balloons to find one containing a prize.
If I were a legal AI wrapper with a valuation somewhere around the GDP of Liechtenstein, I’m not sure I’d invite the optics of a literal AI bubble pop. Indeed, I might steer away from the imagery of empty inflation altogether. But I guess they need to find some way to spend all that money they’ve raised and I guess it’s going straight into that Party City balloon budget.
It’s symbolism that a novelist might dismiss as too heavy handed. Luckily, reality doesn’t have to worry about literary critics.
Harvey is certainly riding high. As is its chief rival Legora, which raised another $550 million this week. But we’re barely more than a month removed from Anthropic crashing the legal tech industry by vibecoding a “Legal” plugin that scans Wikipedia. Anthropic’s new tool didn’t turn out to be ready for primetime, but as a warning shot, it put the legal sector on notice that the big AI players can swoop in and drink this milkshake whenever they decide to get serious about it.
Something’s got to give, right? One theory I’d overlooked, but that might make sense of it all, is that investors might be betting that these companies will get sucked up by OpenAI or Anthropic once they cash in at the stock market. Under this logic, money keeps flowing into these companies because investors see it as backdoor access to the big player gravy train. An intriguing possibility… because if and when one of the leading AI manufacturers wants to get into the legal market, incumbent clients could prove key. But is it billions upon billions of dollars key?
A decade ago, Legalweek — then called Legaltech NY — was derisively called “eDiscovery week” because document review drowned out most other legal tech applications. In 2026, it’s undeniably “AI week” (and there’s a word cloud to prove it). Though after an intense week of AI talk, there aren’t really any answers.
The literal bubble may have popped, but the figurative one still grows tantalizingly larger by the day, raising questions up and down the market.
Many of my interviews talked like betters gearing up for March Madness, except instead of points scored, the over/under line was set at how many of the vendors in the exhibit hall won’t exist in 24 months. Whether a product of consolidation or losing out, most people envision the majority of the exhibit hall roster will be the answer to a trivia question a couple years from now. But what about the foundation of this market? Whenever I would ask about the fact that that hundreds of billions of dollars in value seem to be entirely supported by a closed loop of IOUs between NVIDIA, Oracle, Microsoft, etc. not unlike a 6th grade science textbook explaining the water cycle, the standard response was a nervous chuckle.
And that’s with no one even mentioning that the federal government has declared a holy war against Anthropic and claims — dubiously, but nonetheless — that any company dealing with the government has to cut ties with arguably the best model out there.
On that note, more providers openly talked about using smaller language models, citing cost and improved precision to the task at hand. But one was willing to come right out and say “to control our own destiny.” Even if the provider wasn’t touting small models, the days of “we have an exclusive relationship with [insert AI big dog here]” are gone. Two years ago, that was the flex, but now it reads as a liability. Everyone plays the field… or at least wants customers to know they are capable of playing the field.
Because whatever happens with the top level of the industry, the lawyers are starting to get too dependent upon AI for it to disappear. In the latest installment of the General Counsel Report from Relativity and FTI, GenAI adoption among corporate legal departments jumped from 44 percent to 87 percent in a single year. That’s a Blackberry-level adoption rate.
But Blackberry isn’t around anymore, even if emailing associates with urgent requests at 4 a.m. remains.
In this way, Legalweek was a tale of two AI conferences. Half the conversation excitedly speculated about agent swarms or the promise of vibecoding, while the other half more or less took all the bleeding-edge stuff as beyond their control, instead talking about what makes for a working legal application. It’s a bit of a Braveheart outlook: “they may take our specific workflow tasks, but they will never take… our DATA MOAT!”
Or whatever it is. The catalog of vetted research, or the informed data structuring, or just the human intelligence oversight — there’s something they claim imbues the model with the soul of a lawyer. Or whatever the equivalent of “a soul” is for a lawyer.
Joe Patrice is a senior editor at Above the Law and co-host of Thinking Like A Lawyer. Feel free to email any tips, questions, or comments. Follow him on Twitter or Bluesky if you’re interested in law, politics, and a healthy dose of college sports news. Joe also serves as a Managing Director at RPN Executive Search.
The post Legalweek 2026: The AI Bubble Literally Pops appeared first on Above the Law.
