The Frozen Middle: Insights from the 2026 RSGI AI Impact Report
Why fluent isn't the same as competent.
In the RSGI research into Harvey usage this year, one phrase kept surfacing, unprompted, across dozens of interviews: the "frozen middle." A Canadian law firm partner put it starkly: "I 100% see the frozen middle. You have power users at the junior level and the senior level but there is a hollowing out in the middle."
Reed Smith's Richard Robbins captured the anxiety behind it: a year ago he could stand up and say lots of people were using Harvey. Now, when he says that, he's "almost apologising, because partners say, ‘so what? What are they doing and why does it matter?’"
That question ‘so what?’ is the whole problem. It points to a distinction that matters more than most AI enthusiasts acknowledge: fluency is not competence.
Fluency to Competency
All of RSGI’s research reveal lawyers who are fluent with AI. They can query it, ask it to summarise a case, and get a first draft. It's how the profession went to near-universal usage in about 18 months. But what we have come to learn is that fluency is not the destination, competency is. It’s the difference between holding a conversation in a foreign language and writing a novel in it. It's when a lawyer stops treating Harvey as a smarter search tool and starts treating it, in one interviewee's words, as "a sparring partner." It's building an agent instead of running a query.
One good example of competence is the Setterwalls partner who created a vendor due diligence agent to turn an unprofitable service line into a viable one. Another is Gilbert + Tobin's Caryn Sandler describing Harvey’s agent builder as one of the two features "most responsible for her firm's business model evolution." Competence changes what a team can sell and the order of value it delivers, not just how fast it can work.
“Gilbert + Tobin's Caryn Sandler describes :Harvey:’s agent builder as one of the two features 'most responsible for her firm's business model evolution.'”
Our data puts a number on the gap. Power users — daily, sophisticated users — save 11 hours a week at law firms, against four hours for everyone else. But more than that, power users are changing the way that they practice law. It’s the difference between AI as an efficiency tool and AI as a structural advantage. The frozen middle sits between those two states: by every usage metric "successful," yet stalled in terms of transformation.
The Accelerating Impact of Legal AI
Explore how leading legal team measure the value of AI, including the shift from usage metrics to legal, commercial, and business outcomes.
Why the Middle Freezes
The mechanism is well understood, if under-discussed. As Setterwalls' Philippa von Seth told us, Harvey "didn't come with a problem it was designed to solve." Unlike email, it doesn't slot into an existing habit. It requires a new habit of mind: seeing your own work as a set of processes that can be redesigned, not a queue of tasks to get through faster. That's a bigger ask than learning a new interface. A lawyer running 10 queries a day looks identical to a power user on a login report, until you ask what those ten queries were for.
The Chill is not Just in the Law
This pattern shows up everywhere AI has moved past the pilot stage. Microsoft's 2026 Work Trend Index found that while roughly half of AI-using employees sit in what it calls the "emergent" zone, only about one in 10 strongly agree AI has transformed how work gets done in their organisation, even as C-suite leaders report enterprise-wide impact at three times that rate.
Deloitte's 2026 report, The State of AI in the Enterprise: The Untapped Edge, describes the same disconnect more bluntly: fewer than 60% of workers with AI access use it in their daily workflow, and 84% of organisations haven't redesigned jobs or workflows around it at all. Writing about those findings in a separate Deloitte Insights article, AI adoption to AI adaptation, Mark Gustafson puts it well: "adoption tells you someone opened the door. It tells you nothing about whether they changed how they work, how they think, or what they're capable of on the other side."
None of this is really about the technology. It's also worth being precise about our use of the term, because "frozen middle" already has a life in management speak, where it describes middle managers who obstruct change. The interviewees who used it in the research put a different connotation on it — describing it as lawyers who use AI consistently at a similar level. Unfortunately, unless legal leaders act, the lawyers in the frozen middle, if left alone, may never move out of it.
Defrosting Approaches
The latest RSGI research is consistent on how to drive AI competency, and it isn't generic training. Tom Trujillo, the Chief Innovation and AI Officer at McGuire Woods, said on a recent Harvey webinar, “training is not change management.”
What works are stories. The single most effective change management tool cited across every interview wasn't a metric, it was a specific account of Harvey transforming one matter. It's tone from the top that's genuine and repeated, not a one-off hackathon. And increasingly, it's performance incentives. Firms that put AI fluency into reviews and bonus criteria are measurably further ahead in terms of the legal and commercial outcomes they see.
The firms and legal teams that pull away from the pack in the next 18 months won't be the ones with the highest login counts. They'll be the ones defrosting the frozen middle one story at a time.
—
To hear more about how practitioners are managing the frozen middle, watch our on-demand webinar featuring a discussion of three advanced Harvey users: Tom Trujillo at McGuireWoods LLP, Russell Davies at dentsu and Javi Díez Tomillo at repsol.
Read Javi’s power users blog to find out why, "People Call Me Harvey."
