Most attorneys have an opinion on AI. Fewer have changed how they actually work.

Bryan McWhorter did. A partner at Knobbe Martens focused on prosecution and dispute resolution across software, telecommunications, and emerging technology, he's spent eighteen months rebuilding how he practices, and he shared with us, On the Record,what changed and what didn't.

An unusual entry point

Bryan came to patent law through his college fencing coach, who was also a patent attorney. What stuck wasn't the legal work itself, but a description of the job: every week, a different invention to understand. Bryan still describes his professional strength in those terms: the ability to get to 80% understanding of a new technology fast enough to formulate legal strategy around it, whether the invention in front of him involves robotics, food science, or autonomous vehicles.

That instinct explains why he was an early adopter when generative AI reached the legal profession. His read on the broader hype cycle is blunt: AI was overhyped, then unfairly dismissed once it failed to live up to inflated expectations.

"We're neither out of jobs nor is it useless," as he puts it.

The tools are a very good natural language automation system. They don't replace legal judgment. But a large share of the work in a prosecution practice is repetitive rather than strategic, and that's where he's put them to work.

A rule borrowed from the courtroom

Bryan’s operating discipline for AI has a name he didn't invent for the occasion.

"There's an old idiom that says you should never ask a question in a deposition you don't already know the answer to. I think that's a really good rule to follow when you're asking AI to create content."

In practice, that means he never asks for a tool to generate something in a patent application unless he already knows exactly how he wants it to look.

That's a companion to a risk he names directly: automation bias. Naming a risk this specifically is itself a signal: the safeguard isn't a vague call for "human oversight." It's a defined failure mode he actively checks for, in his own work and in his associates'.

What the tools are actually good at

The clearest illustration Bryan offers isn't about drafting at all. During a due diligence matter, he was handed a patent family of 56 related patents, each with a file history running over 1,300 pages, and asked to check for limiting statements tied to a single term across all of them. He set AI to review the full set overnight. By morning, he had a ranked list of the five most significant limiting statements, information that turned out to be dispositive for the deal.

That's not a generative task, and Bryan is careful to separate it from drafting in his own mental model. It's an automated review task, the kind of needle-in-a-haystack search that would otherwise cost an associate hours, if it got done comprehensively at all.

His drafting discipline shows up in two places. First, in how he builds an application: "I create the bones, it puts the flesh on the bones, I put the skin on top." He builds the framework and the figures himself, lets the tool fill in detailed text against that structure, then edits and finalizes.

Second, in how he runs inventor calls. Where he once relied on shared, implicit understanding with a co-counsel, he now states the invention's focus, flow, and key elements out loud, on the assumption that an AI system is listening and needs the same context a junior colleague would.

The last guild system

Bryan's view of what AI means for training gets more structural.

"The legal industry is maybe the last guild system in the United States. You join and you are an apprentice under a master attorney, and you learn by doing."

AI is changing what that transfer looks like without replacing it. Bryan shares his own style documents with his associates inside the same tools he uses himself. The associate isn't drafting from scratch anymore, they're supervising an AI that is, then bringing him the result. He still runs the same top-level review he always did. The skill being taught hasn't changed. Where it sits in the process has moved up a level.

What this means for firms still deciding

The firms that come out ahead in this transition won't be the ones with the most access to AI tools. Nearly everyone will have access. The gap will be between firms that treat the technology as infrastructure to build a practice around, and firms that treat it as an experiment to run on the side. Bryan's account suggests what the first group looks like in practice: specific rules for when to trust output, a named risk they actively guard against, and a training model that gives junior attorneys a faster path to the same judgment, not a shortcut around building it.

------------------
Watch the full conversation between Bryan McWhorter and Matt Maitland, recorded live as part of DeepIP's On The Record series: replay available.

‍