Nick Berk

Bridging the Law / Tech Divide

Lawyer | Engineer. ex-Microsoft, Cooley (Biglaw), Harvard Law

[Nick Berk @theNickBerk]

https://t.co/UU9qnT60LE

[Nick Berk @theNickBerk]

Yes, "partner" no longer means "equity owner" (and it hasn't for some time). It's misleading, and firms use the title to charge more.

But I don't think this change is leading to worse service (just more expensive service).

"Partner" has never directly meant "better lawyer." It meant "has a book of business." They're not the same.

Many of these "non-equity" partners are "better lawyers" than the equity partners...they just aren't as good as biz dev. I know many senior associates and "of counsel" who were way better lawyers than the partners they worked for, but they weren't as good at wining and dining, etc.

Theoretically "equity owner" would mean that they care more about the firm? But in practice it means they just care more about the profit.

It's hard to judge quality in the law. "Equity owner" has been a proxy...but it's always been imperfect.

[JH @writeclimbrun]

this year, more than 50% of "partners" at big law firms are non-equity.

in other words, more than half of the people with the title are not actually partners. @jborstein

more often than not, if you encounter a partner, they're not.

[Nick Berk @theNickBerk]

“The tool itself is just a pipe from the law to Claude,” Dingman said. “It doesn’t tell Claude how to go analyze a legal issue. Claude’s actually pretty good at it out of the box.”

So much of "AI for legal" is getting the right information in the right format to the LLM. Projects like this are really exciting.

[Bob Ambrogi @bobambrogi]

Two lawyers have built an MCP connector for Claude that they say outperforms the legal research giants. And it is completely free for any lawyer to use. https://t.co/50LjZJfnh0

[Nick Berk @theNickBerk]

For any recurring AI task (drafting a motion, editing a contract, etc.), save the prompt you use as a text file.

Every time you get a draft back, make your edits as normal. Then periodically (weekly, or after ~5-10 examples), feed the AI: (1) your original prompt, and (2) a handful of before/after pairs — the draft it gave you and your edited final version. Ask it to rewrite the prompt to incorporate whatever it's missing.

Here's an example prompt:

"Look across all the edits I made and identify the recurring patterns — things I changed more than once, not one-off fixes. Then rewrite my original prompt to bake those patterns in, so future drafts need less editing. Keep the rest of the prompt's structure and intent intact. Show me the updated prompt, and briefly list what you changed and why."

[Nick Berk @theNickBerk]

I don't really care about the definition of an "AI-Native Law Firm."

The distinction that matters is firms that are static vs. those that are constantly improving (ideally automatically).

A lot of firms will use AI to take a first pass at a draft. But the firms with the real advantage will capture every edit, so the next first draft is better. It's basically backpropagation.

It's not about using AI. It's about codifying, and continuously improving, knowledge and processes. Historically, this has been very difficult (templates and written playbooks were the only way). AI makes it easy - you often just have to update your prompt!

Firms that are constantly improving will grow their advantage over time. They will provide better quality, more consistency, and faster turnaround.

[WillC @willchen500]

Any founders of AI native firms want to chime in here?

[victor @stokebuilder]

wtf kimi just topped fable, gpt, and grok by 2x on Harvey's legal benchmarks? am i reading this right?

[Nick Berk @theNickBerk]

At this point, I think it's clear that there will be legal-specific tools. Lawyers like tools tailored for their workflows, even when the underlying AI is the same. Some firms will go "claude-native," but many will pay someone for a nice layer on top.

The real question is pricing power. Ultimately, these companies have to charge a mark-up on token costs. How much can they charge?

The worry: Harvey/Legora charge a high premium without actually being better - with their margin propped up by switching costs. This is unfortunately the case for a lot of software today.

But switching costs will be much smaller moving forward. AI makes it easy to (1) replicate software; and (2) move data. So firms will switch to cheaper alternatives if they're similar.

Right now, Harvey and Legora (and others) are investing billions into creating the best experience in AI for legal. Given that this is essentially a new product category, there's a ton of low-hanging fruit. I'm thrilled that they're doing the hard and expensive engineering work to figure out how to make AI work best for lawyers.

(And they've actually been subsidizing token costs so far, so it's been a simpler financial decision - though that is quickly changing as they move to consumption-based pricing).

But it's going to get harder to maintain that advantage over time. At some point, the legal capabilities are "good enough" - and it's going to be hard to differentiate.

So it comes down to one thing: can their engineering orgs stay far enough ahead of the copycats to keep charging a significant premium? That's theoretically always been the case for software companies, but in a world of lower switching costs, it's more challenging than ever.

[Zach Abramowitz @ZachAbramowitz]

For those waiting for Harvey, Legora and the rest of the legal AI applications to get eaten by the frontier models, I’ve got bad news for you

[Nick Berk @theNickBerk]

A common trait of great writers and great engineers: they love deleting.

[Luke Ingalls @luke_n_ingalls]

On the one hand slop code is annoying but on the other hand I LOVE deleting code. This is causing lots of tension day-to-day

[Nick Berk @theNickBerk]

https://t.co/cqT5rCjmdq

[Nick Berk @theNickBerk]

It's sad to see the Utah legal regulatory sandbox end. I hope this doesn't become evidence that "regulatory change can't improve legal services."

The sandbox - which allowed non-lawyer ownership of law firms and allowed non-lawyers to practice law - was ultimately narrowed so it became a shell of its initial intention.

The state got cold feet and significantly limited the program after intense lobbying from existing lawyers. Notably, there were very few complaints from consumers (20 complaints out of 117,000 service provided, as of last year).

If you want to learn more, check out the @StanfordCLP's report from last year (in comments).

[Natalie Anne Knowlton @natalalleycat]

ICYMI, the Utah legal regulatory sandbox is officially in sunset mode (Phase 3). It has stopped accepting new applications, and only 7 entities remain. End date: August 14, 2027.

I'd like to say it had a good run but ... 🌆🪦

[Nick Berk @theNickBerk]

Innovating in the law requires a tricky balance between respecting experience - and questioning the way things are done.

That's true in every industry, but in the law experience is especially important.

In engineering, you can be a whiz and compete with much more senior engineers easily.

In the law, there is often no substitute for reps. You can't "reason" your way to the answer. It's about feel.

The companies that can codify and capture that experience, while radically transforming the process, will win.

[Pavin @thefineprintesq]

Lawtwitter (myself included) like to criticize legal AI a lot (for valid reasons), but imo Norm is going the right direction.

Buy books, use AI to eat the belly of the giant whale that is Big Law

[Nick Berk @theNickBerk]

I've been experimenting with when to type v. dictate. I know a lot of people are moving entirely to dictation but I've found that doesn't really work for me.

I've found dictation is great when (a) I know exactly what I want to say, and it would take too long; (b) I have a bunch of jumbled thoughts and i need to make sense of them.

But when I have a good idea of what i want to say, but I'm not completely sure, the editing process that happens during typing is extremely useful.

[Matthew Berman @MatthewBerman]

I think more clearly when I type than when I dictate

[Nick Berk @theNickBerk]

High-end legal work is going to be insulated from change for a long time. It is not a good signal for how disruption is happening.

When it comes to bet-the-company litigation, or a major transaction, people simply do not want to take a chance. They don't think "let's save a few bucks" or "let's try an innovative approach."

[Derek T. Muller @derektmuller]

I've learned that Jones Day is offering bonuses to Supreme Court clerks of $700,000. It appears AI is not yet coming for those positions. @DavidLat

[Nick Berk @theNickBerk]

There's a deeper divide about the general purpose of law school.

A lot of law schools have convinced themselves that they're in the business of providing a more "theoretical/academic" education - as opposed to a practical one.

Most people would be SHOCKED to learn how little practical knowledge law school graduates have. Unless they went out of their way to take "practical" classes, they don't know how to file a case, negotiate a contract, etc.

Law schools are content to have students spend three years "thinking about the law" and leave the rest to industry. When you show up at a firm, they assume you have essentially zero knowledge (this is particularly true for transactional law).

So many decisions law school makes are downstream of this "academic" view. They want to resist seeing law school as a vocational school, and instead see it as a place where people explore ideas.

Modern law is going to require knowing how to use AI. Everyone understands that. But many law schools just don't see it as their job to provide that kind of practical education, even though that's what the majority of students want.

[Sean Silverman: Silverman Bar Exam & LSAT Tutoring @BarExamTutor]

The problem with this policy, as I see it, is that law students have purchased a very expensive product. Have you seen tuition lately?

They are the buyer; the school is the seller. If the purchaser of a product, an adult no less, believes that he/she will get more out of that purchase by having a laptop in class, that should be their right. Adults don't need other adults to tell them what's best for them.

I'd, personally, say the same about AI, but the computer ban is even more nonsensical.

Frankly, this will probably go on until us oldies have died out. When people who have used computers their entire educational life are the ones making these decisions, I don't think computers will be banned from classrooms. But until then, people will, by nature, cling to the past.