How LeanLaw Is Building AI Into Legal Revenue Operations: A Conversation With Our CTO

Most of the AI conversation in legal is about practicing law — research, drafting, contract review. That work matters and it’s moving fast. There’s a second place AI belongs in a law firm that gets far less attention: the operation that turns legal work into collected revenue. We sat down with LeanLaw CTO Fred Willerup to ask what LeanLaw is actually building, what’s live today, what’s early, and where he draws the line.

A note on what this is: a top-layer look at how we think about AI. We’d rather tell you plainly what’s shipping, what’s in beta, and what we’re still figuring out than overclaim — which, as you’ll see, is more or less Fred’s posture throughout.

Is AI on the money more valuable than AI on the law?

Fred: I wouldn’t claim that. AI on the law side is extremely valuable — in some areas it’s reshaping the whole industry, and it’s critical that firms figure out how to use it. I don’t think we need to come in with some bold statement that the money side matters more. What I’d say is that operations and revenue are where we add something. The AI attention has mostly gone to practicing law. The place we can help is the part that’s less talked about — the workflow of taking on a client and actually getting paid, which is really about making the firm a better business.

(We’ve written more about where that line falls between AI that practices law and AI on the firm’s money.)

There are two different ways AI shows up for a firm. How do you think about them?

Fred: There’s embedded AI — AI features inside a product that do something for you there. Our chat function, ClerkAI, is embedded AI: you ask a question in plain language instead of clicking around to find the data. We’re also building an embedded feature to make invoice review easier — if you have 200-plus invoices to send in a month, it could flag the handful you should definitely look at, the handful you probably should, and leave the rest. Once you trust that, invoice review gets much faster.

Then there’s the other one, which I think is the bigger shift: the firm brings its own AI — Claude, or another agentic tool — and that agent needs to work across the firm’s systems. So a firm’s agent could pull information from their intake system and set up the client, the matter, and the rates in LeanLaw, and even kick off a retainer request. We don’t run that AI. Our job is to make sure LeanLaw works cleanly with it.

That second one has a name now — Agent Experience.

Fred: It does, and it’s a useful frame. Agent Experience — AX — is the emerging discipline for designing products so that AI agents acting on a user’s behalf can actually use them. Gartner has it on its 2026 Hype Cycle for Agentic AI.

Worth being precise, because the term gets used two ways. One camp uses AX to mean workforce readiness — training people to work alongside agents. That’s not what I’m talking about. The other camp means product and API design: making your system legible and usable to an agent. That’s the one we’re building for. Internally we just call it being agent-ready.

Isn’t this the same connected-experience argument LeanLaw already makes?

Fred: It’s exactly the same principle, extended. Not one tool that tries to do everything, but one connected experience across the systems a firm already uses — and now that includes the firm’s own AI. The agent gets information out of one system, puts it into LeanLaw, does other things across the stack. We want to be the system that plays well in that setup rather than the one that blocks it.

For a non-technical reader — what’s an MCP, and where are you with it?

Fred: Systems talk to each other through APIs. An MCP is essentially a specialized kind of API that an AI agent has an easier time working with. We have an API an agent can already use, and our MCP layer is in private beta today.

On the standard itself: MCP was donated to the Linux Foundation’s Agentic AI Foundation at the end of 2025 — co-founded by Anthropic, Block, and OpenAI. That matters more than it sounds. It means firms investing in agent workflows aren’t betting on one vendor’s proprietary thing.

We’re moving deliberately here, and it’s in beta on purpose. This is a system that touches billing and trust accounting, where the failure modes are unforgiving — so the guardrails come before the capability. I’d rather work closely with a small number of firms and get it right than ship it wide and find out.

Let’s talk about ClerkAI. What does it actually do today?

Fred: I’ll be straight: it’s early. It’s minimal right now. No client was banging on the door asking for it — we built it because being able to interact with your product in natural language is becoming table stakes. Today it’s a way to ask questions about your data. It doesn’t take actions yet. Where it earns its keep is the questions that are hard to get out of a standard report, when what you’re looking for goes beyond the normal filters.

For example, a firm might want every client who still has open receivables and has paid at least one invoice in the past — so they can send reminders without chasing people who never became paying clients at all. That’s genuinely hard to pull from a normal report. That kind of question is where ClerkAI is useful.

What makes asking ClerkAI different from pasting your numbers into a general chatbot?

Fred: A few things, and I won’t overstate them. It knows the LeanLaw data model, so it’s less likely to make wrong assumptions about your data. It’s grounded in your actual data and what the data means. It keeps you in one place — no screenshotting or exporting your numbers and carrying them somewhere else. And it talks straight to our database, which matters for performance: an outside tool pulling a lot of data through an API is slow by comparison. I wouldn’t claim it’s magic. I’d claim it’s grounded and it stays where your data already lives.

Where is ClerkAI headed?

Fred: A few directions. Turning an answer into a saved custom report you can come back to weekly. Recreating a report or dashboard you like — even from a picture of one you had in a previous system. And eventually letting it do things, not just answer.

That last one we’re deliberately slow on, and I want to be clear about why. The moment AI can take actions inside a system that touches billing and client funds, the guardrails have to be right before the capability ships — not after. So we build the safeguards first and earn that step by step. Today it answers questions; the rest comes when we can stand behind it.

You had an interesting way of framing where this all goes — AI as a user.

Fred: Yeah. We have a product, and people at the firm use it. Increasingly the firm’s AI is going to be one of those users too — essentially taking a seat, like an admin, doing the things people can’t get to or don’t want to do by hand. That’s the shift we’re building for. Not AI as a bolt-on feature, but AI as another user of the system.

That shift shows up in the numbers before it shows up anywhere else — it’s part of how AI changes a firm’s realization rate, and it’s one of the reasons AI is making fixed-fee billing harder to avoid.

The question every firm asks: is our data safe? Do you train AI on it?

Fred: We use third-party AI providers who don’t train on your data without our permission — and we don’t give it. We also don’t sell your data to anyone. Because this data includes privileged client information, we’re deliberate about how it moves through every AI feature we build.

The underlying concern — client-confidential information from a privileged matter surfacing somewhere it shouldn’t — is completely valid, and it’s exactly why we’re careful about how customer data moves through any AI feature we build. With attorney-client privileged information, that caution isn’t optional.

(If you’re working through this question for your own firm, we wrote a longer answer on whether you can put client financial data into an AI tool.)

Where does this go next?

Fred: The near-term focus is making LeanLaw genuinely agent-ready, plus making ClerkAI better and more useful. We think the pull is coming — firms will expect their software to work with their AI the way they already expect integrations and connectors. It’ll be table stakes.

We’re building it now, and this is the point where outside input is worth the most. So here’s a direct ask: if your firm has started building agent workflows — or you’re thinking about how your own AI should work with your billing and accounting data — get in touch. We’re taking a small number of firms into the private beta, and the firms in it will shape what this becomes.

You can see what’s available today, and tell us what you’re building, on the agent-ready LeanLaw page — the public API is available now, and the MCP layer is in private beta.