Back in March, a Legalweek panel featuring Steno surfaced a problem the legal industry hasn't solved: if AI is absorbing the work junior lawyers used to learn on, how do those lawyers develop the judgment that senior lawyers now use to vet AI output?
One speaker on that panel, a technology professional at a major law firm, called it a snake eating its own tail. Today's senior lawyers can catch AI's mistakes and refine its output into something usable. But that skill was built by doing the very work now being handed to a large language model. If the training ground disappears, what happens to the next generation of senior lawyers?
Above the Law picked the thread back up five months later, at ILTACON 2026, where Steno Director of Brand Jeff Cox joined a panel titled "AI as a Teammate: Use Cases That Drive Talent Development." The piece frames the ILTACON panel as a kind of answer to the question Legalweek raised. The panel was moderated by Kathryn Rattigan (Robinson+Cole) and coordinated by TaShika Lewis (Cooley LLP), and also included Damien Riehl (Clio), Daniel Himmel (Legora), and Nick Hafen (BYU Law School).
As Hafen put it, "It's not that there's no value in going through the paper boxes or the diligence room. But once you know what a clause looks like, doing it again and again doesn't add a lot of value to your training. And I think AI can concentrate and enhance that training."
Jeff's contribution built on that idea with a specific proposal. Instead of AI trained to write like a generic associate, what if the tool was trained to write like one specific lawyer, the one a junior attorney would most want to learn from? He calls it the doppelgänger.
"A generic AI writing system produces output calibrated against a broad model of what 'legal writing' looks like," Jeff said. "That can be useful, but it's not what we're talking about here." A doppelgänger works differently. It's built on a narrow, deep model of how one lawyer actually thinks: how they open an argument, how they handle a bad fact, how they concede a point without it reading as retreat.
The mentorship case builds on this idea. Traditional mentorship depends on the right matter, the right assignment, and a partner with bandwidth to give, which makes it a question of access as much as merit.
Jeff proposed an alternative where an associate drafts a summary, runs it against the partner's doppelgänger, studies where their own instincts diverged from the model's feedback, and walks into the real conversation with the partner already holding a theory of what that feedback will be. "It compresses the feedback loop that mentorship at its best is always trying to create," he said.
And because the doppelgänger doesn't have a calendar, it's available to every associate at the firm, not just the ones already working with that particular partner. "It becomes infrastructure," Jeff said, "available without competing for partner time and calendar time."
Building one, he explained, starts with the parts of legal practice that already look inefficient on paper: the redlined draft, the marked-up hard copy, the memo explaining why an argument went one way instead of another. That material is exactly what trains the model. "Why did this argument go first?" Jeff asked. "Why was this fact treated as background rather than foreground? Writing out the reasoning behind instructional choices attached to the work is what transforms the voice replica into a genuine thinking partner."
He closed with the case for urgency, framed against an industry that famously prefers to be second. "The lawyers and law firms who build this now will be practicing at a materially different level than those who wait," Jeff said. "I think the question is not whether to do it. The question is how much longer you can afford to wait."