Why this matters
In March 2026, Harvey raised $200 million at an $11 billion valuation, co-led by GIC and Sequoia. (Harvey, for the record, is named after Harvey Specter from the show Suits.)
Now read what the money is for. Harvey said the funds would scale the embedded legal engineering teams that build and optimise AI agents inside the world's top law firms and legal departments. An "agent" here just means software that runs a task end to end. The company already serves more than 700 customers across 58 countries. It could have poured the cash into research alone. Instead, a large share goes to humans who deploy.
Why?
Because software does not change how a lawyer works. People do.
A legal AI tool can pass every benchmark and still gather dust. The hard part is not building the model. The hard part is the gap between "the AI can do this" and "this firm actually does it on a Tuesday." That gap is where most legal tech quietly dies. Closing it is the whole job. So the company hires people whose only role is to close it.
How a forward deployed engineer actually works
The term was popularised by Palantir, the data company. Palantir sent engineers to live inside client sites and solve problems on the ground, rather than shipping software and hoping. AI companies borrowed the idea. Harvey calls its version legal engineering.
Here is what it looks like in practice.
A forward deployed engineer embeds full time inside a single law firm for six to nine months. Their job is to take the firm's messiest, most valuable work — M&A diligence, securities filings, regulatory memos, document review — and turn it into AI that fits that firm exactly.
The role is really three jobs in one. Part researcher, who learns how the firm works. Part engineer, who builds. Part trainer, who gets people to use it.
The time split is telling. Roughly a third goes to conversation — with partners, associates, knowledge lawyers, and IT. A third goes to plumbing the firm's document system, usually iManage or NetDocuments. A third goes to making the AI accurate and testing it.
Notice the biggest chunk is talking, not coding.
That is because a law firm is not one user. It is many. Partners, associates, and knowledge lawyers each adopt new tools for different reasons. What convinces a partner bores an associate. What an associate loves, a partner distrusts. You cannot script your way past that. You have to sit with people.
What the engineer leaves behind is not a generic chatbot. It is a custom agent. It speaks in the firm's house style. It cites the firm's own precedents. It respects the firm's confidentiality walls. Harvey's customers now run more than 25,000 such custom agents. Each one is a workflow that a human helped shape.
This is also why the best legal engineers are usually former lawyers, not pure coders. The rare skill is legal judgment — knowing what a good output looks like, and spotting where the dangerous edge cases hide. The code is the easy half.
What this means for you
Most firms and in-house teams in India are watching this from the outside. The lesson is not "go raise $11 billion." It is simpler, and more useful.
The reason legal AI works at the top firms is not the model. Anyone can buy a model now. Mike OSS even gives one away for free. The reason it works is the embedded human who translates the firm's real work into something the AI does reliably. That function is the moat.
So the real question for your firm is not "which AI tool should we buy?" It is "who is going to deploy it?"
You probably cannot hire a full-time forward deployed engineer for nine months. Very few teams outside the global elite can. But you do not need to own that function. You can rent it.
That is exactly what legal engineering as a service is. Someone who understands both the law and the tooling comes in. They map your highest-volume workflow. They connect it to your documents. They build the thing. They train your people to trust it. You get the Harvey-style embed without the Harvey-style budget. This is the work aircounsel does.
If you want to start before anyone arrives, do one thing this week. Name your single most repetitive, highest-volume task — the one that eats junior hours and follows the same shape every time. A standard NDA review. First-pass diligence. A recurring compliance memo. That task is your first deployment. Everything else follows from it.
The big legal AI companies have already placed their bet. The edge is not the software. It is the person who makes the software fit.
Now you know why they hire them. The next question is whether you will.