Subscribe
Share
Share
Embed
Ropes & Gray attorneys provide timely analysis on legal developments, court decisions and changes in legislation and regulations.
Transcript:
Ed Black: I want to welcome everybody to the latest installment of the Ropes & Gray alumni podcast. I’m particularly pleased to be here today because I’m here with my friend, former colleague, maybe future colleague—we’ll see what he says—Steven Obiajulu. Steven has led a long life of success. He went to MIT, took his degree in mechanical engineering—from there, on to Harvard Law School. Joined Ropes & Gray first as a technical advisor where he worked in the patent practice, but that wasn’t enough for him, so from there, he moved on and became a very successful attorney in our private equity practice in a group we call the private capital transactions group. And from there, he left the firm to start UseJunior, a legal tech startup. Steven, it’s just great to have you.
Steven Obiajulu: Thanks for having me, Ed. It’s great to be here.
Ed Black: My first question is a question that the introduction emphasizes. You’ve just had this extraordinary professional journey. What took you through all of these phases of your career?
Steven Obiajulu: They’re blessings, and I think they start from very supportive parents who encouraged me to pursue math, science, and engineering, so that’s how I got into MIT. We grew up watching the DARPA Grand Challenge and the self-driving cars that MIT’s John Leonard would make. I think probably around then and maybe seeing the Iron Man movie, I really wanted to go to MIT.
Ed Black: The DARPA Grand Challenge? Wasn’t that in Nevada? Did you physically go, watch the cars?
Steven Obiajulu: No, we didn’t physically go, but it was shown on either the Discovery Channel or the Science Channel. My younger brother and I were avid watchers of those channels, and so we watched those competitions. Then, I go to MIT, and I take “Dynamics and Controls” with John Leonard—he was my professor. Excellent professor, great guy. I did ask him, “AI and self-driving cars seem to be a really cool industry?” And this was 2011, I think. The view then among professors was that—I was 19 years old—it’s kind of too late for someone that age to be getting into AI at that point.
Ed Black: Too late?
Steven Obiajulu: Yes. We had this one very simple small language model back then—it was a bigram model—that I got to do as a class project. But it seemed like I was too late to the party for AI in 2011—it was already too close to being solved. Both my parents are lawyers. My older sister’s a lawyer. Law was just a great profession. I think it’s still a great profession, great time to be a lawyer. Harvard Law is just a great place to go, and Ropes was supportive of that—and then, patent law was excellent. Getting to meet inventors was like getting to meet your heroes—they were excellent people. I learned a ton.
When Ropes spun out its patent prosecution practice, I asked a senior partner who knew me well, “What do you think would be the right practice for me?” And he said, “M&A law.” And so, I followed that counsel. I loved the M&A group at Ropes & Gray—very talented colleagues. The level of precision, attention to detail, and sense of urgency, I don’t know that I’ve seen that really anywhere else. I worked at the Precision Engineering Research Group at MIT, so there was precision there, but the urgency, client focus, client service was really impressive. It was a very difficult decision to leave the M&A group after six years as an associate. The short reason for why I left was I was offered an opportunity. A friend from MIT, who’s now our angel investor, called and we were just talking about oxygen concentration that could help COVID patients—seems like there could be funding for something like that. I kind of joked, “A legal tech that helps lawyers could be an even simpler product to commercialize.” I was asked, “What makes you say that?” And for context, Ropes has been early and responsible adopters of AI technology for some time. It was actually Neill Jakobe who recommended that I use Kira. Now, Kira’s the most widely used AI software among lawyers. It doesn’t get the most press, but if you look at the data, it’s the most widely used. I started using Kira pre-COVID at Neill’s recommendation, and I just thought it was a powerful tool—its accuracy was excellent. I think in five years of using it, I’m not sure I saw a single error. It could have false positives, but it just wouldn’t really miss change of control provisions. If you got halfway decent optical character recognition, the PDFs were good.
Anyway, I had been familiar with legal tech for some years now, but I heard from our former head of knowledge management, Pat Diaz, that there were certain features that weren’t fully there for some tools. I had already been asked to give a presentation in January 2025 on AI tools to the private capital transactions group. Before giving the presentation, I wanted to hear about how the tools work—what works, what doesn’t work—and when the issues were mentioned to me with some of the existing tools, some of them were pretty straightforward to fix. And so, I went ahead and just made a simple prototype to address probably the most glaring issue, and I showed the output to Pat Diaz, and he thought it looked good: on par or maybe better—I won’t put words in his mouth—but maybe better than what we were paying for. It was like two weekends of work to get to that point. We’ve done a lot more work since then. And it just showed me that if a major player in the space had a tool that could be improved like that by someone who had been out of practice for a decade, I just saw that there was opportunity. I relayed that to my friend, and he wanted to invest and wanted to get started right away. He was on vacation and ended the vacation early, on a Saturday. I gave notice on Monday that I was going to work on this project because I see it as something that can improve the tools for lawyers at Ropes & Gray and similar firms. I think lawyers deserve to have good tools, especially if there’s low-hanging fruit that can substantially improve the user experience.
Ed Black: You go through each of these experiences, and through an unexpected confluence of circumstances, legal tech and AI as a possible tool is literally knocking on your door. A weekend comes along where the venture investor, the existing tech solutions, and our own Pat Diaz—who was head of our knowledge management team for many years—are all knocking on your door at the exact same time, all having the same message: that is, “Things could be better.” You grab the opportunity—it sounds great.
Steven Obiajulu: We looked for other CEOs, by the way. I didn’t want to just go and leave Ropes. I really enjoyed being at Ropes. It’s an excellent place to practice law. I’d been there for 11 years. I asked two Harvard schoolmates—one from the business school, one from the law school—but they all declined because it just seemed too technical, it seemed difficult to build. So, I then realized it would have to be me to step into this role.
Ed Black: On the professional journey, that also is a nice segue to talk about where you currently are. Can you give me a little bit about what UseJunior is, what it does just to provide some context?
Steven Obiajulu: UseJunior, the most distinctive thing about it is it’s a document comparison tool. You can send two versions of a document into the tool, and it will send you a red-line (which Microsoft Word can do), but then, by return email, it’ll send you a bullet-point issues list or a table format issues list of the type that maybe a law clerk would help with preparing. So, maybe a three-column table, a header, a brief description of the issue and then the legal implications, or four columns where it has a heading of the issue, how that issue was addressed in the original draft, how it’s addressed in the revised draft, and the legal implications of the change.
Ed Black: The problem that UseJunior solves is this problem of going from a document that’s just landed in your inbox with a client pounding on your door saying, “I want the deal to move forward. What do we make of this thing?” And it takes you from that moment when you have the two documents through a redline compare and an issues list substantive compare, and it does that through an email exchange?
Steven Obiajulu: Exactly. The problem that it is set up to solve is that it’s 11:30 p.m., you just got opposing counsel’s draft, and the client wants to know that night how bad their markup is. How much risk was introduced? Why can’t we just sign this already? Why are the lawyers holding this up? And so, you can highlight the risks that were introduced by the markup very quickly, within maybe five or 10 minutes, or at least have it in your view. There should be humans in the loop—this is not a replacement for lawyers. This is a tool that runs asynchronously in parallel while you do your own review. But, yes, that’s the problem that it’s meant to address.
Ed Black: Agreements vary quite a bit. Here at Ropes & Gray, if I just think about what crosses my desk, it’s everything from one-page agreements covering things like NDAs and material transfer arrangements and other things all the way up to some of the agreements we do for credit financing—300 pages, 400 pages. Are there scoping or subject matter limits, or does UseJunior just do this for everything?
Steven Obiajulu: To give credit where it’s due, a lot of this is the result of Google Gemini doing a really good job training their model—it has the whole internet, including a lot of legal training materials, in its in-model memory. It’ll do a decent job in a number of areas, and it has a lot of versatility. Now, you could ask then, “Why not just go directly to Google Gemini?” I actually think for things like summarization that actually does make sense, but with comparison, there are unique problems that have to be solved to unhobble the ability that is present in the model already. I see our approach as allowing Gemini to work the way it should have been able to work when comparing documents, but currently, if you just go put a raw comparison into Gemini, its intelligence is limited in comparison relative to its ability on summarization. And this is no knock on the Gemini product. We had the pleasure of meeting with Google’s head of enterprise AI, Crispin Velez, and one of his engineers, and we presented UseJunior to them. They’re supportive. They’re an excellent team. We’ve got a ton of respect for them. They’re aware of some of these things, and I’m sure they’re working on solutions, but that’s why our tool is so versatile.
Ed Black: Because you’ve got the power of Google standing behind it.
Steven Obiajulu: There’s the power of Google, but then there’s also a very specific unhobbling that we have to do in our pre-processing. Now, they say with open houses that the neighbors are the first people to show up, and I think there are probably other providers who would very much like to use this same approach, but they don’t currently. I’m not going to go into a lot of technical details. It’s a substantial difference, and I encourage people to try our tool side by side on document comparison, and they’ll notice a difference.
Ed Black: How do you manage getting that functionality, which does leverage that powerful underlying tool in the right way, while maintaining security, confidentiality, keeping everything in the right place, in the right track, without the wrong person finding out the wrong thing at any point?
Steven Obiajulu: UseJunior is ISO 27001-certified and, more importantly, is SOC 2 Type II-certified over a three-month monitoring period. For those who may not be familiar with the technical standards, ISO is an international standard, and then SOC 2 is the American Institute of Certified Public Accountants, an American standard, which is a little bit more stringent and more technical. SOC 2 is distinctive in that it’s not just looking at whether you have the policies in place. They monitor you for three months, placing scanners on your servers, monitoring your network operations to confirm that you actually comply with your policies—that you behave as you advertise. We have gotten both of those certifications with unqualified opinions, so by a certified public accountant in the case of SOC 2 because that’s the one that involves a certified public accountant. We’ve had auditors go through, and they’re thorough. Kind of like how the bar exam is the big examination for being a lawyer, SOC 2 is like that for a software organization, and we went through that and passed it. Trust.usejunior.com goes into the dozens of controls that we have in place that are still scanned and updated.
Then also, Google’s been very supportive. They have a prompt-logging exception in place with us, so they don’t even log our prompts—it’s like a zero-data-retention arrangement. You might say, “Isn’t that what they do with everyone, or isn’t that what they do with all the enterprise users?” All enterprise-paying users, it used to be that that was the case, but now, you have to get a prompt-logging exception with them, which I encourage all law firms to do. If they’re using Google Gemini, if you don’t already have it, I would say get a prompt-logging exception. So, we have that. And then, I personally have gone through every step of the processing pipeline at the level of the individual machines, the individual processing units, and I’ve verified where the data goes. We soft delete every five minutes, we hard delete every 45 minutes, we hard delete the chat history every 24 hours, and we hard delete all the attachments every 48 hours.
Ed Black: Let me shift gears a little bit and ask some bigger contextual questions. You described your personal professional story, and your jump to the tech startup was because you, with a couple of weekends of work, were able to prove to yourself and to an investor that the world could be different in certain ways and that AI tools really could change things. To expand on that, how do you see the whole AI toolset affecting the practice of law?
Steven Obiajulu: I think that these tools are going to be good for clients, and I think it’s going to be good for lawyers. I even think it’ll be good training for young lawyers. It’s not my opinion, but if you talk with the researchers—I don’t really know too many people at OpenAI, but I do know some people at Gemini and some folks at some of these other labs—it seems like the general agreement is that—and they say this publicly, if you care to look, you’ll find it—there’s this paper that many of them reference, Measuring AI Ability to Complete Long Tasks by METR, and the upshot of that paper is that there’s a Moore’s Law-like effect with respect to AI autonomy. About every seven months, the AI’s ability to operate autonomously doubles. So, if the AI can operate autonomously for 30 seconds today—which is roughly where I’d put it—then next year this time it’s a minute. That’s kind of how people think about it, and all my conclusions about how AI’s going to develop is based on that trajectory.
A lot of people talk about efficiency and cost savings. And it’s not just me—the CEO of Box, who really gets this, and he’s been saying that people are going to use AI for quality assurance. The quality is already amazingly high at Ropes & Gray. Instead of seeing this as, “We’re going to cut costs. Who can we lay off with this?” another way to look at it is, “You’ve got this unit of intelligence. How can I use this unit of intelligence to assure the quality of the output, to make it such that errors never happen?” If you run commercially available, consumer-grade AI with the extended reasoning against some of the largest transactions that have occurred over the past few years and you say, “Were there any errors in it?” it’ll pick those things up in five minutes. And so, why not run that check before it goes to a client or goes and gets published on EDGAR forever? I think it’s for sure going to be used to bring a legal work product that’s already at an amazingly high level, it’s going to be an impeccable level, at least along certain dimensions.
Ed Black: ChatGPT sort of broke the glass ceiling for AI—AI tech has been around forever—but until the fall of 2023, people didn’t really envision its application in legal services. Then, ChatGPT comes along with a very splashy public unveiling, and within a couple of months, people are trying to take the bar exam with ChatGPT, saying that the bot is passing. Goldman Sachs publishes a study—since it’s been published it’s been substantially discredited, but at the time, everyone thought it was brilliant—about how AI will replace 40-50% of legal tasks. And all of that initial thinking, it was all about replacement. That is, we’re going to be doing everything the same way we’re doing it now, it’s just if there are 10 things to do and now 10 humans do it for 10 billable hours, then when AI comes along, four to five of those hours will be done by machines and presumably that will either be free to clients or a much lower cost to clients, and so, it’ll be replacing a certain percentage of human errors with machine errors. Yet, when I think about what you just pointed out, these deep capabilities that allow you to see the work differently and to put it on a different footing in terms of its accuracy and what it delivers, that isn’t replacing a single hour of human work— that’s transforming the legal work product by taking it to the next level. If you had to pick replacement and we just fire a bunch of attorneys versus transformation, where we have the same number of attorneys but they’re all doing things differently, which of those do you think AI will be: replacement or transformation?
Steven Obiajulu: I think for leading firms it’s going to be transformation. I’m already seeing that—I see that we’re able to get more done for more users. So, it’s not even theoretical—we’re already seeing that.
Ed Black: What do you think the future holds for UseJunior? How do you think your company and your product move forward from here?
Steven Obiajulu: We want the tool to help lawyers provide excellent quality work to their clients, and so, the job’s finished when you get the level of assistance from the Junior AI as you would expect from a helpful, fairly resourceful junior associate or law clerk. The junior associates keep their jobs, and then they control it and they have more leverage at the firms that adopt technologies like this, and they can grow their market share. So, what does it look like for us? Currently, the tasks that Junior does are like 30-second tasks that are on loop. One feature is the document comparison. Comparing two versions of a document, you can break that up into steps so that each step is like a 30-second task and then it can be coherent. Summarization’s one thing where it’s not really bounded by that 30-second limit—just based on my experience. We do tools like that.
But why couldn’t you say, “Can you search on iManage?” We’ve spoken with iManage about this—we’re waiting to hear back—but you search iManage, you go through a bunch of documents, and this assistant reads and summarizes all of them, concludes which ones are most relevant, and sends you the iManage links back by return email, and says, “Here’s what I’ve seen,” like you would with a law clerk. One Turing test that you could create for this is, what if you can’t tell the difference whether or not you emailed it to a helpful law clerk or if you emailed it to our tool? If it goes off, searches iManage, comes back with results, searches the internet if necessary—usually it’s not necessary to do internet search in our work—what are the tells? We run it and, “How come I can tell it was an AI?” “Well, because there’s this, this, and that.” “Alright. What patch can you do?” We’re at the limit of the models in some places, but that’s where I see it going.
Ed Black: It’s interesting that when you think about that down-the-road product, the root functionality that’s leveraged in your assignment to your junior clerk is it starts with the functionality that exists today, which is this take a series of things and compares them and understands the differences not only in a redline way but in a more meaningful, substantive way. Once you can start making those comparisons, then you can put longer workflows in place where we’re making comparisons and judgments among multiple documents, not just two; across a larger database; and maybe with more substantive range—addressing problems that take many more steps, like you would go to a junior clerk and say, “I’ve got this kind of a deal that has this kind of issue. Take a look at my document management system and find me things that you think are relevant to helping me understand this kind of deal with this kind of problem.” That could be a multifaceted comparison, and yet, it all starts with that core ability to compare and understand, which is quite interesting.
Steven Obiajulu: Our comparison is the most distinctive thing I think that would justify a purchase by itself. It runs the gamut—it can do summarization. It can fall back to Google Gemini and give you all the things that Google Gemini can. Plus, my favorite feature of the tool is actually this term sheet check, which was actually recommended by a Ropes & Gray partner, Sam Levitt. It was an excellent idea—and we implemented it in like 48 hours—a bullet point list from an email, and you take a contract, and you send it into the tool. The tool takes the bullet point list, creates a checklist for itself—a to-do list—and it goes through the to-do list one at a time and it checks, “Is that term present in the contract?” And it has to write down a score. If it doesn’t write down a score, it’s sent back, and it has to go through the loop again for that particular question. And then, you get a checklist showing that it checked off compliance on all these things. It’s substantially better accuracy than if you just send one prompt in. It focuses all its attention on the one task at a time. So, it’s creating its own workflow, and it executes its workflow live on the backend, it gives you, I think, the highest-accuracy comparison for a term sheet check out of any AI tool. That’s just a feature we include, by the way—we don’t even really talk about it that much.
Ed Black: The deliverable back to the client then is a report, an output, that tells you, “Does the agreement live up to the term sheet or not?” And I imagine, since it’s digital, that could be in any number of formats—is that right?
Steven Obiajulu: Yes. We were at a negotiation one time where I had drafted a commercial agreement and I had actually missed one term in the term sheet. I sent it to the term sheet checker with the counterparty copied, and then the AI said, “You didn’t implement this term sheet term.” And then, the counterparty said, “Is that true?” and I said, “Yes. Sorry, my apologies.” It established goodwill— it wasn’t intentional because why would I have copied you on that communication with the AI? It creates these interesting dynamics where there’s more trust in the AI than in the user of it.
Ed Black: Very interesting. I want to ask some questions that are not about your legal career, not about your current product or the future of legal tech in AI, but just about you. One of the things that I know, because we’ve been colleagues along the way for many years, is that in addition to being an engineer, of course, you’re a scholar of thought—you’ve studied philosophy. Is there a philosopher that you’ve studied where something that philosopher has said or done really captures your view of how you see the world?
Steven Obiajulu: Thomas Aquinas is probably the philosopher whom I really draw a lot of light from. I really like his scholastic method, where he presents objections before presenting his argument and tries to present the three strongest forms of the people who would object to him. I think that is really admirable, and I like that approach.
Ed Black: The core philosophical skill is an ability to see the other side and to be self-critical. What a wonderful choice. One more question, and this is what I call the deep values question: In a peanut butter and jelly sandwich, what is more important—the peanut butter or the jelly?
Steven Obiajulu: I would probably have to say the peanut butter. Maybe it’s pretty contextual, but I think high protein/low sugar is probably important.
Ed Black: Peanut butter is at least the nutritional driver, if not in some larger sense, the life force driver, of a peanut butter and jelly sandwich. By the way, if anyone listens to this and they want to reach out to you, how do people find you?
Steven Obiajulu: They can find me on LinkedIn: Steven Obiajulu. And you can follow us on LinkedIn: UseJunior, for weekly updates.
Ed Black: Steven, thank you so much. It’s been a pleasure to connect with all of our alumni, and I’m sure the alums will all be very, very interested. For all our alumni out there, thank you for joining us for the latest edition of the Ropes & Gray alumni podcast. Please visit our website, alumni.ropesgray.com, to stay up to date with all alumni events and to get the latest news about our firm and our lawyers. This podcast will be up on the Ropes & Gray website, but it’s also available wherever you find your podcasts. Thanks so much for listening in.