Inside RA Capital’s AI-Powered Approach

Banner artwork by AI Generator / Shuttterstock.com

My firm, RA Capital Management, was the most active life science investor last year. In 2024, we closed a financing transaction every 2.5 days on average; if you do the same math for the Confidential Disclosure Agreements we enter into in order to conduct due diligence on possible investments, it’s three per day. We have close to 200 employees, which includes our incubator that forms and supports NewCos. We typically raise at least one new investment vehicle every year, and have multiple fund strategies.

In sum, to use a non-legal term of art, we have a shi*load of work. And we handle as much of it as possible in-house — everything from private investment in public equity (PIPE) transactions, to fund formation, to employment issues, and vendor contracts. Our legal team is lean (but not mean!): a general counsel, a deputy general counsel, a general counsel to our incubator, and currently, to help us keep up with the volume of work, a first-year associate from a large law firm on a training rotation with us, and a fifth-year attorney from another large law firm who is on temporary secondment. 

To manage this load with our small team we must find every opportunity to routinize, standardize, and commoditize our legal work, and use every efficiency-enhancing tool at our disposal. We are proud to operate like a factory, and not an atelier. 

Commoditizing routine legal agreements — or, “AI, avant la lettre”

The greatest inefficiency for any lawyer is negotiating the same type of agreement — over, and over, and over, and over… For the kind of work this legal team does, the solution most often lies with creating and promoting “Model Documents” for the types of agreements we use most frequently, and in taking the lead to ensure wide adoption in our ecosystem. This is because, fundamentally, almost all of what we do involves a counterparty with whom we must negotiate our documents. We can’t and won’t simply jam “our form” down their throats. However, there is a lot of room between doing that, and dealing with each transaction (of the same type) as an entirely new one, where the documents are fully bespoke and we must scrutinize every word anew. 

After reviewing his fifth vendor contract in his second week of work here, our first-year associate secondee wistfully asked: “Do you think someday their AI will send over a contract, our AI will review and comment on it, and then our AIs would agree on a compromise version that was down the middle and fair to both parties?” Today of course, no lawyer would be willing to do that — the AI tools have not reached a place where we have that much faith in them. 

What we have tried to do instead is to create a human version of that process, by forming coalitions representing the relevant stakeholders to agree on a down-the-middle version of standard documents that we frequently use, so that we are starting from the consensus version each time. That drastically reduces the red ink on each document and the number of times we send it back and forth. Here are the agreements that we deal with most often, and how we have, or are attempting to, “commoditize” them:

Using AI as our junior associate to prepare the first draft of legal documents and memos

We routinely use Claude and ChatGPT (so-called “sandboxed” enterprise versions, of course) when we need to draft a bespoke document (e.g., a Collaboration Agreement or Side Letter). Given our decades of collective experience in the RA Capital Legal Department, it is pretty hard to come up with any type of legal document for which we don’t already have a precedent — so sure, we could use that. And we will refer to that as well. But the point is: you should always be seeing what AI serves up as an alternative — for so many reasons: 1) it is better to have more precedent examples; 2) even if it is 90 percent garbage, there may be a useful gem in there; 3) AI is like our first-year associate, and will get better every year — we need to stay abreast of its progress, so we can know when and how it can best help us. 

We use AI — but do not exclusively rely on it! — for legal questions, for all the same reasons. Our incubator GC uses it, e.g., for research on non-compete law in different jurisdictions where we may be considering hires. Of course, she will then check Claude’s work with the underlying source materials! 

And yes, AI can do the first drafts of huge sections of your fund Private Placement Memorandum. Go ahead and roll your eyes at some of the boneheaded mistakes it made! First, yes, we all still need constant reaffirmation that we are smarter than ChatGPT4.0, so enjoy that while it lasts. Second, for that first draft, would you rather use a) AI or b) a second-year associate? If you answered b), did you know they bill almost US$1,000/hour????

Using AI tools designed for lawyers

We are constantly asking our peers what they like and use, and demoing products ourselves. Not gonna lie — most of them just seem to be wrappers around ChatGPT so far. We have licensed one AI tool to assist us with reviewing and redlining our CDAs and vendor contracts, because in both of those cases we have a well-established “play book” that we feed into the AI to guide its review. We are not going to name the vendor, as we don’t have enough experience under our belt yet to be in a place to endorse it — and so far, John Henry seems to be maintaining an edge over the drilling machine. That is, it is not yet clear that the tool is more efficient than our humans. And so yes, for now it often adds in another layer — kind of like asking a lawyer whom you believe is somewhat incompetent to mark up a vendor agreement, and then having to review and redo a lot of their work. So what. Suck it up. The point is the same as we made above: if you don’t take the time to train that agent now, it will never get better. But if you put in some time now, one day, just like that very green attorney, it will grow up and actually become incredibly useful to you!

Using AI to optimize interns

Every fall and spring semester, we have a law school intern, and we often have summer interns as well. The usual hesitation around hiring interns is not misplaced: will they take up more of our time than they free up, because they are here for such a short time and are total rookies? We have always viewed our intern program as part of giving back to the legal community by training junior lawyers and so have not let that concern hold us back. However, for the first time, with this summer’s intern, we have seen how AI can significantly assist novice workers. AI can vastly improve the ratio of “time spent explaining stuff to the intern” to “value-add contributions from the intern,” quickly overcoming their inexperience and organizational knowledge gaps. The particular AI tool that has had the biggest impact in that respect is Glean, a platform that collects all of the knowledge and information from across our organization, whatever the source, and enables the user to query that combined information in one place. So as our summer intern sits in on a meeting, I see her asking Glean questions like “what is a DDQ?” and then “who at RA Capital can help me fill in the fund performance information for this DDQ?” She is then able to turn to the task right after the call, with very little need for further explanation.

Other automation tools

Our team uses Smartsheet to input requests for vendor contracts and term sheets. Given the volume of vendor contracts that we process, having the business lead answer all the key questions upfront vastly expedites the legal team’s review. For example, how critical is this vendor to our operations? Will sensitive data be shared outside of our walls? This tool also allows us to track the status of these contracts as they move through our system. In the case of our term sheets, that tool allows us to capture our deal terms on the back end so that we have a searchable data set to query (e.g., “how often do we use participating preferred stock?”) and also provides embedded guidance for the deal lead on each of the terms (e.g., what is “market” or “RA standard”).

Old-fashioned checklists

One of the thought leaders in our industry wrote an entire “Manifesto” about checklists. They’re the only way to absolutely ensure consistency and completeness and avoid sloppy errors. When we first engage with outside counsel, we give them a document we call our “Cookbook” about how to engage with us — SOPs, signature blocks, deal terms that are non-negotiable, and contact information for the various team members they will need to interact with from finance, operations, and legal. 

Given the necessary secrecy surrounding such transactions, we often have precious little time to review documents by the time that we learn that one of our portfolio companies is actively engaged in an M&A process. Accordingly, we have an M&A checklist to ensure we don’t miss any issues that are critical to our firm when we are reviewing lengthy and complex documents late at night on a short deadline.

With the turnover in our legal team (we take a new first-year secondee every year and interns from BU Law School for both the spring and fall semesters) , it has become essential to develop a checklist of our desired/best practice deal terms; the “oral tradition” breaks down as sous chefs are added to the kitchen.

Building your own AI agents

At RA Capital, all employees have been encouraged to play around with building AI Agents for tedious tasks we perform that can be better tackled with an AI assistant. If that sounds scary, read this recent New York Times article, and go play around with one of the websites they mention. Your agent might be as simple as something that turns those tedious (I mean highly informative!) law firm update memos you get into a condensed podcast for you to listen to on your ride home. There is an AI app for that … maybe you’re using it right now to listen to two AI podcasters discuss this article.