Legal Isn’t a Service Anymore — It’s Becoming Infrastructure (Brian Elliott, Scale LLP / 5.4 Technologies)
Show Notes
Brian Elliott, partner at Scale LLP and founder of 5.4 Technologies, breaks down a shift most of the market is still misreading. This isn’t about lawyers getting faster with AI tools. It’s about legal work being decomposed into systems that can execute without lawyers in the loop.
Inside an 80-attorney, fully remote firm operating across 21 states, Brian is actively encoding legal judgment into reusable “skills” and deploying them across the organization. The result is a real-world test of what happens when a profession built on bespoke expertise starts behaving like infrastructure. Adoption is uneven—not because the tech doesn’t work, but because incentives don’t align. When your value is tied to billable time, turning your judgment into a system compresses your own leverage.
The conversation moves past surface-level automation and into where value is actually collapsing. Roughly 80% of legal work—research, drafting, document review—is already machine-executable. The remaining 20% is where lawyers still matter: prioritization, risk calibration, and strategic sequencing. But even that layer is being tested. Brian argues that what lawyers call “judgment” is ultimately pattern matching across prior outcomes, and that those patterns can be encoded, scaled, and improved beyond human limits.
The failure mode shows up clearly in current tools. AI can flag 30 issues in a simple $20,000 contract—but a competent lawyer knows that level of scrutiny destroys the economics of the deal. The gap isn’t intelligence. It’s proportionality. The next frontier isn’t better detection—it’s context-aware decision systems that understand when not to act.
On the client side, the shift is already underway. Companies are pulling work in-house, using AI to handle the majority of legal workflows and bringing in lawyers only for edge cases. One client delivers a 19-page AI-generated estate plan analysis before the lawyer even starts. That flips the model: the lawyer is no longer the origin point of analysis, but the validator of it.
Brian’s longer-term vision is agent-to-agent legal infrastructure. Systems detect issues, propose solutions, and, when needed, interface directly with law firm systems to resolve them—without humans managing the process step-by-step. Legal work becomes asynchronous oversight rather than synchronous execution.
What’s unresolved is liability and trust. The current system is built on human accountability. When decisions are made by encoded frameworks, responsibility becomes diffuse. That’s the constraint slowing full adoption—not capability.
The bottom line is simple. Legal is moving from a profession organized around individuals to a system organized around decision architectures. Firms that don’t transition will not just lose efficiency—they’ll lose their position in the workflow entirely.
Topics Covered
Why “legal as infrastructure” changes where value lives
The real 80/20 split between automation and human judgment
Encoding legal strategy vs. assisting it
Client-side AI and the collapse of the traditional firm funnel
Agent-to-agent transactions and removing humans from execution loops
Liability, regulation, and the real bottlenecks to full automation
What replaces the junior associate pipeline
by Jason Todd Wade / BackTier / NinjaAI - AI Visibility - SEO, GEO, AEO
Transcript
Brian Elliott Introduces 5.4 Technologies and AI's Role in Legal Practice
Host: Good, good, I press record like right before I pick up. So if we need to edit, we can, but otherwise we're good to go. Awesome. So I pulled up — you have linked from your Pod Match: Elliott.law. But you're also part of Scale LLP.
Brian Elliott: Elliott.law is my personal branding site. I do all of my legal work through Scale. So I'm a partner at Scale LLP, but this gives me sort of my outlet for all the things that I like to do. I also founded — last year, I founded a tech company called 5.4 Technologies, where we are developing AI tools for law practice.
So right now, we're doing the kinds of things that people are asking for. Basically, we're walking down — if you saw the latest article that I put on Elliott.law, which is the full-stack AI law firm, that's the road map that we're walking our clients down.
But right now, we're doing two things. We're doing skills on the client side and the law firm side. We're developing agents for client side and law firm side. And then we're going to be building out the infrastructure that's going to connect those two with a transaction layer. So it will be agent-to-agent legal transactions.
Host: I like the nuts and bolts. I'm currently pro se in a family situation, so I've been dealing with a whole lot of technology, which has made me really good at it. But I'm curious about the process — and I can share mine — of like the entire process of when you get something. A lot of times opposing counsel will do a tracking email, which I'm sure just goes into the system, the database. So I always take the number off, I always take the email off. It probably doesn't do anything, and they'll have it in the subject too. I take it off. I'm not making their job easier.
But from the very beginning — either somebody files something or they get a contract — how do you start on these?
Brian Elliott: So right now, we don't. It's a full human interface layer. Traditional law firms will start cases the same way, which is usually with a phone call, an engagement letter, and then you open a file. And that's the way it will be for a while. We've got some regulatory issues that are going to require us to do that.
Now, the technology is there that we could do an API on the law firm side that just sends a packet and says, here's the information we need. The client's agent fills out and supplies the information, and then we can just do that engagement letter process to start the attorney-client relationship in an automated way. Nobody's doing that right now, as far as I know, because we've got regulatory issues about the attorney-client relationship and how that needs to happen.
So we're trying not to leapfrog that. That's certainly where we think it's going to go in the next 12 to 18 months. But right now, what we're trying to do — what I'm trying to do for my own practice and what I'm helping people do with 5.4 Technologies — is to get it ready.
So that means on the law firm side, we need to look at our information as discrete areas of judgment that can be encoded. When you get a family law packet, a family law file, we've got our templates that we know we always use and go to. Everybody's situation is going to be different. Well, AI is really good at matching a set of data to a set of templates and making that work. So what used to cost five or ten or fifteen thousand dollars could probably now be done in minutes through AI.
So there's a value question there. Is it still worth fifteen thousand dollars? I don't know. Maybe it is, because you've got the human lawyer behind it certifying it, saying that this is good, and it's my malpractice license that's on the line if it isn't. So there's some value there. Anyway, we're working through those types of issues with law firms across the board.
How AI Can Encode Legal Judgment and Surpass Human Limits
Brian Elliott: I'll give you an example. We just rolled out — I did this for Scale, I'll give that example because I can talk about them, that's my firm — we spent some time distilling actual legal judgment into Claude skills that will run automatically. And we rolled that across an 80-person law firm.
The kinds of conversations we're having internally at Scale are the kinds you would expect. Some attorneys embracing it and saying, “This is great, it'll help me do my work in less time.” Fine. And others are saying, “Well, let's slow down a minute. I'm not sure I'm signing off on this judgment that this Claude skill is putting out. How do I intervene in that process? What's the next step?”
I think we're a little bit early still for full end-to-end automation. You hear it all the time — there are AI-native law firms that are doing this. But I feel like I've spoken to a lot of people who are trying to. There's a lot of bondo, suspenders, and duct tape behind the scenes. Ostensibly it's an AI-native law firm, but what that really means today, in April of 2026, is they're using encoded skills and agents to process probably 80% of the commodity work, but 20% of the work still needs human intervention and human judgment. And overall, there's this whole regulatory backdrop we have to contend with.
Host: What I noticed in my experience — and I've had wins and I've had losses, of course — is the AI is really good at pulling case law. Very, very good. And then you always have to double-check it. But it hasn't been wrong once. It's gotten a lot better since Trump's lawyers with the judge and the funny cases. Especially if you check GPT, then Claude, and end up — and Get Law. Get Law's like a Nazi about stuff, oh my gosh.
But where lawyers have the benefits — paralegals are screwed. For discovery, ripping all the text — actually, GPT has gotten pretty good at OCR. I have another one that I like better, where I could do a corpus and then dump it in a large context window like Claude. And then for discovery, you could — I did it just the other day. I said, “Where did the judge say this? Where did they say this?” And I said, “Give me all the context surrounding it,” and it instantly did it. It's not like the OCR with Adobe where you can search a term but then get 20 PDFs.
But what it's bad at, as you know, is procedure, is strategy. It's just horrible. It has led me astray. I don't trust it anymore. You guys are worth every dime — the good ones, the top 10%.
Brian Elliott: That's exactly what I'm working on at 5.4 Technologies. Look, we're not trying to compete with the bulk of legal tech and the frontier models — we're not doing that. But what I do think is that it's possible to encode that part of the strategy, the legal judgment. I think that's reducible to a set of instructions.
I've been testing it. I've got a couple of models that I'm working with, and I'm getting really good results. So far, I'm just testing it on my own clients. I'll run a real-life scenario, but before I run it, I'll have what I would do. And then I'll run it through my model. And I get pretty good results, and it's getting better.
When you look at the way people say — you hear a lot in the legal field that AI doesn't do judgment, AI can't strategize, that it's good at all the other stuff but it can't do that — I disagree. Because when we do it as lawyers, what I'm doing as a human lawyer, I'm taking a set of facts in front of me, this data that has come to me from a client's case, and I'm applying that in my brain anecdotally to the cases that I've seen like this before, and the archetypes of those outcomes that I've seen before. But that's a limited data set. That's only my experience.
Maybe I can cross-reference that with other people at my firm or cases I've read about or cases I have access to. But I can't hold all of the thousands of cases and their outcomes in my brain. It's not possible. So can we get to a stage where AI can do that level of advanced legal reasoning better than a human? I think it's obvious. I think it will do it. It's just a matter of putting the right guardrails around it and not letting it jump to conclusions or skip steps in the process. And that's where I think it's really exciting. We're not there yet, but that's what I'm working on.
Why Current AI Tools Miss the Mark on Legal Context and Proportionality
Host: With contracts — there's the famous one of a few months ago where somebody took a 600-page contract and found the one thing that was going to screw them. I've taken over 2,000 documents, just ripped it, OCR'd it. I've taken almost a 400-page PDF into GPT. And I just said, “What is this?” And it tells me.
That's what I think lawyers could benefit the most from. I'm not a lawyer, and I'm not too friendly with some lawyers — I'm the worst person to speak about lawyer-client relationship. But they can — their paralegals — every lawyer should tell their paralegals, “I'm paying for GPT. It's twenty a month. I pay eight. I'm paying for Claude. You need to use it as much as possible. Double-check everything you do — just start double-checking.”
And I think for lawyers, because they have so much on their minds — and you guys are clearly intelligent — but there's only so much you can hold in your head. Whereas with AI, think of it as a super-fancy to-do list. That's how I do it. I don't go hunting. Whenever I'm saying, “Write an email to the judicial assistant saying I just filed this with the clerk,” and then I think through a few other things, which actually for context are going to make it better. I'd say, “Well, what about this? What about this?” And you can kind of think through it — not an annoying weird chat thing — and then I say, “Do the JA thing again.” And then I don't have to look for it. And this made it a little bit better, or it's made it more careful.
That's what I think people could almost use it as — a thinking partner and a sparring partner. And to give different opinions. Some people I know can't understand me. What I always do is I think things through three ways: the way I want them, the way somebody else wants them, and the way it'll probably be — God's will, whatever you want to call it. And I think that's where AI can help you with scenarios.
I think GPT is the best. Claude will sometimes mess it up or get too verbose. Gemini is just weird, even though it's good, looking at formatted stuff. And Get Law — do you mess with it?
Brian Elliott: Yeah. How do they make money? I don't know. I have a big question mark on a lot of how people make money, but we'll see. It's a good model if it lasts.
Let me give you the opposite of some of the things that I've seen. We have great tools — some branded legal tech tools that'll do contract markups. So this is great. You put it through the contract markup tool, and it finds all the issues. But here's one of the issues that came up with me a couple weeks ago.
I had an attorney at my firm use one of these tools, and they marked up a contract. But it was a marketing agreement — a 2½-page marketing agreement that had a total dollar value of $20,000. A 90-day, $20,000 marketing agreement, 2½ pages. Well, the tool is really good. It found 30 issues in 2½ pages. But I can't send that to my client. I can't say, “Hey, look how great we are, we found 30 issues,” because it's not worth it. By the time I negotiate 30 issues, I'm going to spend as much money on this contract as it takes to mark it up.
So in that context, the tool did a very poor job of understanding the scope and the risk of what was at risk. This is a 90-day, mutually-terminable-at-will, $20,000 contract. Let's give it a light pass, just find out if there are any big risk areas, and if there's not, approve it, go on, just get it done.
That's where I think the AI tools are missing the mark. They all seem to want to flex their muscles with how much they can deliver and how thorough the output can be. And oftentimes our job is to make it smaller, shorter, do less, and deliver faster results by telling the client, “We don't need to worry about this, just sign it and move on.”
Clients Use AI to Redefine the Lawyer's Role from Originator to Validator
Host: Yeah, what kind of company would do that much? They'll spend that much in two days on AdWords. And they probably had you sign an NDA too.
Brian Elliott: In the legal market right now, there's a lot of energy around making these tools work and implementing them into the workflows that we already have. My thesis is that the future of legal practice isn't lawyers using good AI tools to make them better at the things we do today. It's going to be clients using AI tools to cut the lawyer out of the process. And I think that's where lawyers need to really understand that we see it.
I got an email this morning from a client — great client, been with us for many years, big company, big national brand. They're informing us that they're taking more of what we normally do in-house, and they're going to do it themselves. They're saying, “Thanks for the relationship, it's been great, you guys did a great job, but we don't need you anymore for this slice.” And I expect in a couple of months, they're going to take another slice, and then another slice. That's just going to erode what we do — which is fine, as long as we understand that's the future. And we need to be building for the future where the clients are going to do 80% of the work themselves.
Even pro se — you're able to handle these huge cases and massive data sets and distill it down to what you need, or the things that you then need to call the lawyer and say, “Okay, I only need you for this slice. I don't need you for the whole thing. Don't bill me for all that, I've already got it. I understand where we're going. But you give me the strategy, get me over the finish line.” I think that's going to be more common for how lawyers work with clients in the future.
Host: I think so too. AI is just going to weed out the weeds. The chaff is still going to be there. You still need lawyers to look at this and say, “No, this strategy is wrong. We need to hold off on this.”
For my friend, he was going through a similar situation to me, and I had a lawyer who was just not good, looking back. So I said to him, “Don't make the mistake.” I said to GPT — my friend, I explained the situation. I do it by voice, I never use prompts. I mean, unless I create them. I can't remember the last time I did an off-the-shelf one. But I said, “Give him advice.” And it said, “You want the top 20% of lawyers” — of course, with what you do, you want the top 5%, right? Top 2%. “And ask them these 10 questions. Are you going to get back to me in 24 to 48 hours?”
If somebody doesn't — people don't like this, especially lawyers — I get reminded when people don't get back with me. But also at the end of the day, if somebody doesn't get back to me — paralegals, what the hell? They don't like it. You're charging your client extra money. Respond to the fucking emails, then. They hate that. You're supposed to give them a few days. But anyways, it was like, “If they say, well, we want to be nice, we want to get along — don't hire them.” It was amazing.
I think you will have much more informed customers. And some people will not like that. Some people, like you, are tech-forward. They will say, “You know what, you're right. I know the top 5% of lawyers in my thing. I'm not going to badmouth them, but now I know.” So having an informed, educated customer — that's going to make your life easier.
Brian Elliott: Right. So one client right now who I connect with on Slack — I'm embedded with them, they're with me all day. The expectation is an almost real-time response. I'm busy during the day, it's not going to be always real-time, but it'll be close. Certainly before the end of the day, we're in constant communication.
But the other day, he sent me an estate plan that another lawyer drew up. It's a set of documents for his whole family, complex estate structure. And he wanted me to give him a second opinion on it. “What does this look like? Is it missing anything?” But before he did, he ran it through whatever system he had, and he gave me the 19-page AI analysis and said, “Here's my starting point. I've already done this work. Now you tell me if the AI got it right or not.” It was just brilliant. I love it.
Because that shortens the work I need to do. I don't need to do all the cognitive work of trudging through all these documents. He's done that already. He gave me the summary. Now I need to double-check it and make sure it did it right and didn't miss anything. So I'm getting halfway there.
But I think a lot of lawyers might recoil at that. They'd say, “Oh, I don't want to look at AI output. I'm going to do it myself. I'm going to start all over from the beginning, and I'll give you my expert judgment.” And that's fine too. For now, there's going to be some element of the client market that's going to want that. They're going to want the 5% best lawyer in the globe to look at this from scratch and develop a very bespoke, customized approach to whatever company issue they're having at the time. I get it. But for most of what lawyers do day-to-day, it's going to be commoditized, it's going to be automated. Let's just get on with that.
Host: Yeah, I'll use folders all the time — projects, whatever they want to call it. Projects in GPT, skills here, gems here. And then it keeps perpetual or running memory — what is the term, I'll remember in a second. They could dump in whatever document — they could dump in his analysis, especially in GPT, which has gotten so good. It could tell you a lot of stuff. So that way you could use your time effectively.
Two things. So I looked on your website — almost everybody on the Scale side has good hair. Have you noticed that? Almost everybody, especially you, have good hair. So I don't know what's going on at Scale, but something simple. It's in the water. I'm so jealous when I see old ladies all the time with this thick gray hair, and I'm like, come on, man. It's on your mother's side. My mother says, “You know, my dad had a full head of hair until he died.” I'm like, come on.
So — are you watching the Y Combinator law stuff? There's several in there.
Brian Elliott's Vision for Agent-to-Agent Legal Transactions and Infrastructure
Brian Elliott: Y Combinator — I am keeping a close watch on the entire market. And here's what I see a lot. Obviously Harvey is taking the headlines for all the money. But there are two directions that legal tech AI is going in the legal market.
One is tools for lawyers to do the things that lawyers do. And the other one is the commoditized, automated — it's like LegalZoom on steroids. They're going to take off the bottom layer.
What I don't see is a lot of money or attention being put on taking the lawyer out of the loop — the tools that should be deployed on the client side that clients can use themselves to give themselves more agency in the process. Because there's not a lot of money there. What you would do, if you're designing that way, you're basically saying that lawyers are going to earn less because the clients are going to do more themselves. And that's not where the tech is going. The tech is going to entrench lawyers in their legal practices more.
I think the right example is the Y Combinator thesis: don't develop tools for lawyers, launch an AI-native law firm. That's their thing. And that's great. But an AI-native law firm is really one of two things. It's either a traditional law firm with AI tools that make it more efficient, or it's a lawyer-coder who's developing agents and things like that to scale what they do across more things.
I think both approaches are missing the real answer, which is: we need to automate, we need to do agent-to-agent legal transactions, and get humans on both sides out of the way.
So if I have, for example, an agent running on a client side that is looking at their communications and throwing up regulatory flags and saying, “Hey, here's a customer service communication, and we're making promises that can't be met that the FTC is going to have an issue with” — today, that's a customer service thing that gets bubbled up to in-house counsel. Then in-house counsel evaluates it. And then if it's important, they call outside counsel. Outside counsel evaluates it. Then there's a meeting two weeks from next Tuesday, and they come up with a plan of how to address it.
Or we can have an agent running that throws up a flag and says, “Here's this issue, here's the solution to the issue.” And if we need outside counsel, it goes out to api@scalefirm.com and says, “I'm going to connect with Scale firm.” Scale's going to develop a solution for it, pass it back. We're going to implement it. And then at the end of the day, the same day, the general counsel gets a report saying, “This problem was surfaced, it was resolved, here's how we resolved it, and we used this outside law firm to do it.” And then done. So you don't have to have a human on both sides of that context.
Host: I've got a securities lawyer. He was at the SEC for a long time, and then he was at the attorney general's. He's been doing securities law for 26 years, I believe. Very good. He's flat-rate. So I don't — it's not my world. I don't know which form. He charges like $35,000 to do a form. He's a great guy, and he's been doing it so long. He said his nickname at the SEC was Gestapo, which — and he's this little Jewish guy. He's like — it was odd. And I was like, “I could see that, Fred.”
So when I started helping him, I said he's going to email me or contact me like 30 times a day. And he did, between texting and stuff, whenever we're working together, collaborating. Which I kind of appreciate, because then I can fix stuff in real time. I got another client — I have to stop doing that. It's like change of scope every 30 seconds. Frank just does it — Fred just doesn't understand it, and that's okay.
So the point was: he said his client checked with a Big Law firm — $400,000, four with. So imagine — with his knowledge, which is essentially like the AI — he could do it in no time. He's probably done not 10, 10,000.
Brian Elliott: That's the thing. There are going to be ways to compete in the future. And I think they're all going to come down to branding around the intelligence and the judgment that you bring to the table. It used to be like lawyers used to brand around the process — “We will take this case for you, and we'll take it from point A to point B, and we're going to try to get the best result.” Right? But really, the process can be automated. So what do you have left? What's the distinguishing factor?
At the end of the day, the area we compete on is going to be: do you want Brian's intelligence applied to it? Do you want Fred's intelligence applied to it? Who are we going to put into this process to bring that judgment to the AI that's actually going to take it all the steps through?
Exploring Scale LLP's Remote, Efficient, and Client-Focused Legal Service Model
Host: Yeah. Well, I wanted to touch on this. So my attorney for a short period was — I had Divorced.law. So it's not bad. I've been buying and selling domain names for a long time. I usually do .com, but the firm has scalefirm.com, which is really good. And then Elliott.law is very good, because you couldn't get Elliott.com anyway. But who did your site?
Brian Elliott: They did a good job. My wife did it. She's my partner in all things. So it's great.
Host: Tell her she did a great job. Holy cow. Maybe besides my client Fred — no, this is the best lawyer site, because lawyers have horrible sites. She did a good job.
Brian Elliott: It's Ghost. It's a content management system. They're competing with Substack. I used to publish on Substack, but Substack's got all kinds of problems with AI visibility and search engine visibility and things like that. So we ported everything out to Ghost.
Host: Wow. Yeah, well, it's exciting. I was happy to see that. And how do you guys at Scale find lawyers? That's their full-time practice, right? They're not — you know.
Brian Elliott: Yeah. Scale is a full-service, full-time legal practice. We've got 80 attorneys in 21 states, and we're fully remote. All of our attorneys work from their own houses or their personal offices or wherever they want to work in the world.
What we do is we're able to pass along savings because we didn't build out real estate infrastructure. We're able to pass along those savings, give you top-tier legal services at a mid-tier price. That's the promise of Scale.
We started in 2018. It was a group of Silicon Valley GCs who knew that there was a different way to approach the market. I started with Scale in 2020, middle of the pandemic, looking for my next thing. I was going to start my own firm. I found Scale. It looked like exactly the kind of work that I wanted to do with the people I wanted to do it with. And it's been amazing growth for us over the last five or six years.
What we do is we look for the best attorneys available anywhere they are, because it doesn't matter to our clients whether we've got a top attorney sitting in South Carolina or Kansas or Seattle. We can service them anywhere. Our main practice areas are corporate, securities, litigation, and intellectual property. Those three main buckets cover probably 80% of what most businesses need on a day-to-day basis for legal services. And we're doing a great job.
Host: With the IP stuff, do you do domain?
Brian Elliott: We do have people who specialize in domain names. We do domain name disputes all the time. We negotiate buying and selling of domain names. Often it comes up in the broader context of M&A, where we've got companies being bought and sold, and the IP portion of it — our IP lawyers will jump in for that portion.
Host: Cool. Yeah, great.
Brian Elliott: It's been great. Well, Jason, look, glad to meet you. I like that people are thinking about AI visibility — that's what we're trying to do as well.
Host: Great. Awesome. You'll do well. All right. Well, thank you. Take care.