DISCO’s 2026 Legal AI Report opens by announcing that this “may be the year the legal tech dam finally burst.” Compared to the legal industry’s baseline, that makes some sense, with simple adoption taking a huge leap this year. Despite the gushing imagery though, the rest of the report suggests less of a violent burst, and more of a substantial, yet controlled opening of the sluice gates at Flood Control Dam #3, allowing a managed release to gently flow downstream.
The survey, run by Ari Kaplan Advisors, drew 104 respondents roughly split between firms and legal departments, and supplemented by direct interviews. Fifty-eight percent of respondents say they’ve folded AI into routine processes, up from 35 percent last year. Seventy-two percent say they’re confident using it for document review over manual techniques, up from 53 percent. And security fears among firm respondents fell from 68 percent to 43 percent.
But what do those numbers really mean? “Routine processes” provide efficiency opportunities, but it’s hardly the same as handing hefty chunks of complex tasks to AI. Using AI for document review? You have been for years! One respondent claimed they would “never go back to a manual review for first- or second-pass review,” but when was the last time anyone cracked open a 2TB drive and started a manual review? We’ve trusted the machines to handle eDiscovery tasks ever since we fought over seed sets. That’s not a knock on the advancements in this space, rather a reminder that this is a slice of the practice that traditionally luddite lawyers have long shown a willingness to hand to the machines.
And when we say security fears are down, is that because the lawyers have learned to trust the AI… or trust their external and internal limits on how it’s used?
Generally speaking, the use cases look like the ones we predicted: deposition summaries, chronologies, first drafts, research, a “sparring partner in litigation.” One participant described the shift as moving “from doing the work to directing the work,” which is… how law firms work. There’s just a robot doing the first-year’s tasks. Which is a win for the lawyers. But the AI evangelists — and more to the point, their investors — need AI to replace human work across the board so they can fulfill their misanthropic vision of harvesting white collar workers like the human batteries from The Matrix while melting poor people for organic pet shampoo. For them, an industry responsibly adding AI to speed its workflow by 11 percent amounts to a waking nightmare leaving them a trillion in the hole.
In last year’s edition, 32 percent of participants reported already realizing savings from AI and 55 percent expected to going forward. That’s nice. Would be a shame if a technology hemorrhaging money with no coherent business model came along to disrupt that! This year, respondents started grumbling — unprompted — about the shift from subscriptions to token and credit pricing. “The token/credit costs seem to be changing as the definition of a task evolves,” one law firm participant said.
Behold a pale horse: and his name that sat on him was Claude, and per token pricing followed with him.
Welcome to Tokengeddon. It was all fun and games while the labs hoped to entrench the product into workflows, but now it’s time to try to make some money. We’re going to find out exactly how much the firm hates letting a summer associate take a first stab at that summary.
One described it as an effort to drive cost savings and efficiency. “Leaders want to do more with less and see legal AI as a means to that end, so they want us to leverage it, and now that cost structures are shifting to per-use tokens, we are beginning to bill back token costs to the business units,” the participant said. “There is a reckoning taking place now, so the use cases will become better defined; I don’t need to burn down the rainforest to find my favorite Chinese restaurant,” the individual added. “There is intense pressure, mainly from executive leadership; using Legal AI to be more efficient is a pillar that helps dilute the perception of legal as a cost center,” added a peer.
But what if it’s not actually saving clients money? Early reports suggest it’s not bringing down hours. For all the drudgery it can alleviate, it also offers insights that lawyers might not have seen before and end up with more time spent creating a better product than a human alone would have produced. That should be good news for the client, but not if they imagine AI as a “get out of bill free” card.
The report also notes that 62 percent of respondents are using AI agents. As an agentic doubter, that number surprised me, but the interviews undermined the top line number. Most of those using agents reported using basic information gathering, scheduled lit hold reports, and bots to “extract data to populate automated forms.” Hardly the soup-to-nuts automation that the hype imagines. Only 10 percent thought legal AI would become exclusively agentic — and frankly that feels too high. Assuming agentic AI ever develops enough reliability to autonomously perform a string of complex legal tasks, the problem becomes a fundamentally human one: keeping the process slow enough for the human brain to exercise meaningful judgment and oversight. So far, lawyers seem to understand that risk.
But the task that agents excel at is “burning massive piles of unnecessary tokens.” At the risk of oversimplification, an agent is a self-managing cascade of prompts that plans, calls a tool, reads the result, checks itself, and loops, burning somewhere between five and 30 times the tokens. Great use cases exist, but the “promise” of agentic AI is that it can take the three prompts a human might need to achieve a working result and replace them with six subagents constantly spinning to produce the same thing. The connection to Tokengeddon isn’t explicit, but it’s there. Because, with mounting open weight competition, “cost per token” isn’t going to get more expensive, so everyone needs to pad those tokens per project numbers.
Again, I’m not saying that there aren’t solid use cases for agents. But it’s going to be interesting to see how these cost concerns collide with agents.
The good news of the report is that more of the legal industry is adopting AI. The better news is that the industry appears to be keeping a tight lid on how it’s deployed. Which is good news for the public sitting underneath that dam.
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.
The post New Report Finds More Lawyers Use AI, But For What Tasks? appeared first on Above the Law.
DISCO’s 2026 Legal AI Report opens by announcing that this “may be the year the legal tech dam finally burst.” Compared to the legal industry’s baseline, that makes some sense, with simple adoption taking a huge leap this year. Despite the gushing imagery though, the rest of the report suggests less of a violent burst, and more of a substantial, yet controlled opening of the sluice gates at Flood Control Dam #3, allowing a managed release to gently flow downstream.
The survey, run by Ari Kaplan Advisors, drew 104 respondents roughly split between firms and legal departments, and supplemented by direct interviews. Fifty-eight percent of respondents say they’ve folded AI into routine processes, up from 35 percent last year. Seventy-two percent say they’re confident using it for document review over manual techniques, up from 53 percent. And security fears among firm respondents fell from 68 percent to 43 percent.
But what do those numbers really mean? “Routine processes” provide efficiency opportunities, but it’s hardly the same as handing hefty chunks of complex tasks to AI. Using AI for document review? You have been for years! One respondent claimed they would “never go back to a manual review for first- or second-pass review,” but when was the last time anyone cracked open a 2TB drive and started a manual review? We’ve trusted the machines to handle eDiscovery tasks ever since we fought over seed sets. That’s not a knock on the advancements in this space, rather a reminder that this is a slice of the practice that traditionally luddite lawyers have long shown a willingness to hand to the machines.
And when we say security fears are down, is that because the lawyers have learned to trust the AI… or trust their external and internal limits on how it’s used?
Generally speaking, the use cases look like the ones we predicted: deposition summaries, chronologies, first drafts, research, a “sparring partner in litigation.” One participant described the shift as moving “from doing the work to directing the work,” which is… how law firms work. There’s just a robot doing the first-year’s tasks. Which is a win for the lawyers. But the AI evangelists — and more to the point, their investors — need AI to replace human work across the board so they can fulfill their misanthropic vision of harvesting white collar workers like the human batteries from The Matrix while melting poor people for organic pet shampoo. For them, an industry responsibly adding AI to speed its workflow by 11 percent amounts to a waking nightmare leaving them a trillion in the hole.
In last year’s edition, 32 percent of participants reported already realizing savings from AI and 55 percent expected to going forward. That’s nice. Would be a shame if a technology hemorrhaging money with no coherent business model came along to disrupt that! This year, respondents started grumbling — unprompted — about the shift from subscriptions to token and credit pricing. “The token/credit costs seem to be changing as the definition of a task evolves,” one law firm participant said.
Behold a pale horse: and his name that sat on him was Claude, and per token pricing followed with him.
Welcome to Tokengeddon. It was all fun and games while the labs hoped to entrench the product into workflows, but now it’s time to try to make some money. We’re going to find out exactly how much the firm hates letting a summer associate take a first stab at that summary.
One described it as an effort to drive cost savings and efficiency. “Leaders want to do more with less and see legal AI as a means to that end, so they want us to leverage it, and now that cost structures are shifting to per-use tokens, we are beginning to bill back token costs to the business units,” the participant said. “There is a reckoning taking place now, so the use cases will become better defined; I don’t need to burn down the rainforest to find my favorite Chinese restaurant,” the individual added. “There is intense pressure, mainly from executive leadership; using Legal AI to be more efficient is a pillar that helps dilute the perception of legal as a cost center,” added a peer.
But what if it’s not actually saving clients money? Early reports suggest it’s not bringing down hours. For all the drudgery it can alleviate, it also offers insights that lawyers might not have seen before and end up with more time spent creating a better product than a human alone would have produced. That should be good news for the client, but not if they imagine AI as a “get out of bill free” card.
The report also notes that 62 percent of respondents are using AI agents. As an agentic doubter, that number surprised me, but the interviews undermined the top line number. Most of those using agents reported using basic information gathering, scheduled lit hold reports, and bots to “extract data to populate automated forms.” Hardly the soup-to-nuts automation that the hype imagines. Only 10 percent thought legal AI would become exclusively agentic — and frankly that feels too high. Assuming agentic AI ever develops enough reliability to autonomously perform a string of complex legal tasks, the problem becomes a fundamentally human one: keeping the process slow enough for the human brain to exercise meaningful judgment and oversight. So far, lawyers seem to understand that risk.
But the task that agents excel at is “burning massive piles of unnecessary tokens.” At the risk of oversimplification, an agent is a self-managing cascade of prompts that plans, calls a tool, reads the result, checks itself, and loops, burning somewhere between five and 30 times the tokens. Great use cases exist, but the “promise” of agentic AI is that it can take the three prompts a human might need to achieve a working result and replace them with six subagents constantly spinning to produce the same thing. The connection to Tokengeddon isn’t explicit, but it’s there. Because, with mounting open weight competition, “cost per token” isn’t going to get more expensive, so everyone needs to pad those tokens per project numbers.
Again, I’m not saying that there aren’t solid use cases for agents. But it’s going to be interesting to see how these cost concerns collide with agents.
The good news of the report is that more of the legal industry is adopting AI. The better news is that the industry appears to be keeping a tight lid on how it’s deployed. Which is good news for the public sitting underneath that dam.
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.
