Why LegalAgent was founded as a law firm for the generative-AI era
Hello, this is Legal Agent.
LegalAgent is a generative-AI startup working to change corporate legal practice through the combination of AI and lawyers. We do not stop at building and selling an AI tool as SaaS. Our own lawyers use an AI agent we built ourselves to deliver the legal work itself, contract review, drafting and more, directly to clients, as an AI-BPO (BPaaS) business.
Concretely, the service is built around: reviews starting from one document per ¥10,000, delivery within one business day, and next-generation legal outsourcing built to move at the speed of the business.
Legal Agent, a law firm for the generative-AI era
Here I want to talk plainly about why I founded a law firm run as an AI-BPO, how generative AI gets used inside it, and what our own product looks like. The founding grew out of real discomfort with a structural problem in the legal industry.
The state of the legal industry
The legal industry has stayed a genuinely legacy environment for a long time. I began my career at Anderson Mori & Tomotsune, one of Japan's largest firms, working on M&A matters, and later moved to AZX Partners, a firm focused on startup legal work. Both firms were excellent, and I learned a great deal at each.
Even so, there was a structural problem common to both: a pyramid where junior associates and staff grind through a huge volume of documents while a senior lawyer corrects it. On a large M&A due diligence project, it is not unusual for a team of ten to fifteen people to work through the night reading materials for days on end. "How long can this way of working go on?" That question sat with everyone on the associate side, myself included.
Clients see it the same way, roughly as "expensive, slow, conservative."
Expensive. Lawyer time bills at tens of thousands of yen per hour, and a single project can involve more than ten lawyers.
Slow. A week for a contract review is common, often a real bottleneck against the pace of the business.
Conservative. Firms flag the risk, but rarely go as far as "so what should we actually do," the business judgment the client actually needs.
What clients want is legal advice that moves at the speed of the business, at a reasonable price, and constructively: closer to helping the business move forward than to eliminating every last risk. But the cost of the pyramid, the delay of routing everything through multiple people, and a conservative, risk-averse posture kept getting in the way. I kept looking for a way to actually solve that.
The shock of generative AI
Then GPT-4o arrived. The moment I tried it, it felt nothing like GPT-3.5. It was a genuinely different kind of tool. I became convinced this would change the industry: on raw processing speed, it was already matching, and in places surpassing, a good associate.
With generative AI, the pyramid built around a large headcount stops being necessary. A small number of lawyers who fully use AI could match a large firm's output at ten times the speed and roughly half the price: a kind of legal service that has not existed before.
Finally, I could build the legal service I actually believed in.
That is what led me to start what is now Legal Agent. Here I want to walk through why we run this as a law firm rather than a SaaS company, and what the development behind it actually looks like, failures included.
Chapter 1: Why BPaaS (a law firm), not SaaS
Handing over a tool alone leaves people stuck
Contract-review AI sold as SaaS has become common recently. It is useful, and it is good for the industry to move forward on this kind of DX. But I think there is a "last mile" that a SaaS model cannot really close: whoever uses the tool is still the one who has to make the final edit and own the responsibility for it.
It genuinely helps when AI flags "there is a risk here." But "so how should this actually be fixed," "how do we negotiate this with the other side," "is this something we can live with as company policy," all of that judgment still falls on a human. What a busy legal team often really wants is not another tool, but someone who can just take it from here.
A large gap in market size
There is also a strategic reason. Japan's legal-tech SaaS market is estimated at roughly ¥20–50 billion, while the market for legal services actually delivered by lawyers is around ¥1.2 trillion.
Competing purely as a SaaS tool means fighting for share of a market worth tens of billions of yen. Taking on the work itself, as a player in that ¥1.2 trillion market, opens a market tens of times larger. The business potential of BPaaS looked far greater than SaaS: use AI to go after share of a trillion-yen market.
An AI at "80 points," corrected by a professional up to "100."
That is why we chose not to stop at offering a tool, but to have our own lawyers use our own AI tooling to deliver a finished product: a BPaaS (Business Process as a Service) model.
BPaaS, in short: where SaaS delivers the software itself, BPaaS delivers the business process, the work, built on top of it.
We do plan to offer the Word add-in we built in-house, our "Cursor for Word," as SaaS to outside lawyers and legal departments as well. But our core business remains the service delivered by professionals who actually know how to use that tool.
For the client, using it is simple: message on Slack or Teams, "please check this contract; the other side is a large company this time, so keep the tone measured." Behind the scenes, AI and lawyers work together, and the result comes back in the same Word file the client already uses.
"Change nothing about how the client already works."
That is what we care about most. No new login, no prompt syntax to learn. Ask in the same chat you already use, and get back the same kind of deliverable you already expect. I believe that is what practical DX actually looks like.
Chapter 2: Behind the operation, the development and the failures
So how is AI actually used behind the scenes? At first, we assumed handing everything to AI would just produce a perfect answer. That turned out to be a real mistake.
The failure: 200 Dify workflows, and hitting a wall
Early on, determined to fully automate the workflow, we built heavily in Dify, a no-code tool. We built out an "NDA flow" and similar processes, with branching logic like "if clause A exists, go to B; if not, go to C."
Before long, we had close to 200 of these workflows, some with dozens of branches and hundreds of checkpoints. And then maintenance simply could not keep up.
A law changes. An interpretation shifts. A particular matter turns out to need an exception. Updating 200 workflows every time that happens is not realistic. Legal practice is a continuous stream of exceptions, and trying to hard-code "every exception" in advance was the wrong approach from the start.
The fix: building our own "Cursor for Word"
So we changed direction. Instead of chasing full automation, we focused on human-in-the-loop: a person working through the task with AI, in dialogue, toward the goal.
Engineers will know Cursor: chat with it while looking at your code, and it produces a diff.
Cursor, briefly: an AI-powered editor for engineers. You instruct it in chat and it revises the code automatically.
"If Word had a Cursor, legal work would change completely."
That thought led us to build, in-house, a Cursor-like add-in that runs inside Microsoft Word, something like having a capable associate permanently on call inside the document. On screen, an AI chat panel sits alongside the Word file.
A lawyer chats with it much like instructing an associate: "revise the definition of confidential information to cover all information." The AI reads the context, edits the contract text directly, and leaves the edit as a tracked change. It can add comments, and reply to comments, too.
The Word add-in screen
That shifted the lawyer's role toward reviewing a draft AI has already written. In practice, we have seen productivity rise five to ten times.
Chapter 3: The organization changes, from pyramid to inverted pyramid
Making full use of generative AI also changes a firm's organizational structure. The old model was a pyramid: a large base of junior lawyers doing the research, a veteran correcting it. The new model is an inverted pyramid: AI drafts, and a veteran judges it instantly.
Roughly ninety percent of the traditional groundwork, research, first drafts, now goes to AI. What that requires instead is an experienced lawyer who can judge, at a glance, whether what AI produced is right.
I think this is a real opportunity for solo practitioners and small firms.
Large matters that used to require headcount to take on become workable for a small team with this "Cursor for Word" and the operating model around it. We are building toward becoming a top global firm for the generative-AI era, through a roll-up that brings together specialized lawyers as a federation, powered by this system and this way of operating.
In closing: redefining professional services
BPaaS and AI-BPO tend to get framed as a cost-cutting story, but I do not think that is the whole point. Something bigger is happening.
A fundamental change in cost structure, an innovation in organizational structure, and through both, a change in what legal service itself actually is.
That is a shift that runs through the structure of the industry itself, and I suspect the same is true well beyond law, for accountants, consultants and engineers alike. Not "hand over a tool and be done," but "work alongside AI and deliver the best possible result." I expect that AI-BPO / BPaaS model to become the standard for professional services going forward.
Building the top global firm of the generative-AI era.
At Legal Agent, we are looking for people who want to build that firm together: lawyers, engineers, legal clerks, sales, corporate roles and more. If any of this speaks to you, we would love to talk. Feel free to reach out for a casual conversation or apply directly.