Mike OSS, Harvey, Legora, and the $80k Question.

Mike OSS, Harvey, Legora, and the $80k Question.

A former Latham associate named Will Chen spent two weeks last month rebuilding the core of Harvey and Legora and put it on GitHub for free. He calls it Mike. The repo crossed a thousand stars in three days. Artificial Lawyer interviewed him. Legal IT Insider ran the line that's been circulating around my feed ever since: Mike doesn't kill Harvey or Legora, but it changes the negotiation. Once a working open source alternative is sitting on GitHub, the renewal-meeting question stops being "is this magic?" and starts being "what exactly am I paying enterprise prices for?"

That second question is an interesting one.

Here's what nobody disputes anymore. Harvey runs on Claude, OpenAI, and Google. Legora runs mostly on Claude. Anthropic's own legal plugin, shipped in April, does native contract review and NDA triage from inside Claude Cowork, free if you have a Pro plan. The intelligence inside the most expensive legal AI products in the market is a frontier model that any law firm can also license directly for twenty or thirty dollars a seat. Will Chen is right about that. The Anthropic launch is what made it impossible to keep pretending otherwise. Mike just put it on GitHub.

So what's left to sell? Workflow integration. DMS connectors. Audit trails. Playbooks. Ethical walls. Governance at scale across hundreds of lawyers. The system around the model.That's real. I want to say this clearly, because this is the part the takedowns get wrong. For an AmLaw 100 firm with eight hundred attorneys spread across iManage, NetDocuments, twenty jurisdictions, and an innovation team that already exists, the orchestration layer is most of what you're buying. Harvey says it's running twenty five thousand custom agents across its customer base. That's not a Claude wrapper. That's a system. At that scale, paying thousands of dollars per seat per month is a defensible number, and Sequoia did the math and decided it was worth eleven billion in March.

Where it gets interesting is the ninety percent of the market that isn't AmLaw 100.

A sixty-lawyer firm in Atlanta or Manchester is not running twenty five thousand agents. They are running on a mixed DMS, a flat-fee billing model on at least half their work, no dedicated IT team, and an AI policy that, statistically, doesn't exist. That last one isn't a guess. The Clio 2025 Legal Trends Report says fifty three percent of legal professionals work at firms with no AI policy or are unaware of one. AllRize's 2025 report on AI-using firms is starker. Half provide zero strategic guidance to lawyers using AI. Fewer than one in five have trained anyone on how to use it. Roughly one in seven have any process for reviewing what the AI produces.

Read those numbers next to an enterprise legal AI contract, often structured around minimum seats and annual commitments, which can run into hundreds of thousands dollars. For a firm that hasn't written down what good looks like, doesn't have anyone to teach prompting, and won't review the output systematically anyway.

The LinkedIn post on this debate had a line about this that I keep coming back to. Buying Harvey or Legora before doing the basic policy and training work is like buying Salesforce before you've figured out your sales process. I'd extend it. Buying Mike is like buying the Salesforce schema and being told to build the rest of Salesforce yourself. The license is free. The work isn't.

I want to be careful here, because I'm in this industry, and the position I'm about to take is the one I'm employed to take.

Harvey would say I'm cherry-picking. They'd argue, fairly, that integrated platforms beat patchwork at every size, and that small firms need integration more, not less. Will Chen would say that's exactly why he built Mike, and that the right answer for SMB firms is to own their stack. Richard Tromans, in his Law Punx episode last September, just said the part most people won't: a lot of firms buying legal AI right now are buying it because of FOMO, and most of them aren't measuring ROI in dollars.

Tromans’ point is the one that bothers me most, because it doesn't matter how clean the SMB argument is if the actual buying behaviour at sixty-lawyer firms is "we need to be seen doing something about AI." A logo on the procurement summary solves that problem. A sober conversation about playbooks doesn't.

So here's where I land, with the appropriate amount of doubt.

The choice that's getting framed as Harvey-or-Mike isn't the right binary for SMB firms. The honest binary is this: do you buy capability before you've built the process, or do you build the process and shape the capability around it? Mike answers the first question by making the capability free, but you still own the build. Harvey and Legora answer it by selling you a finished system designed for someone twenty times your size. Anthropic's plugin hands you the assistant and tells you to figure out the rest yourself.

What's missing from the debate is that the SMB firm equation isn't a downsized version of the AmLaw equation. It's a different equation entirely. The DMS is unorganised and the billing is flat-fee on half the work, which means margin matters more than associate leverage. The IT function is a partner and an outside Managed Service Provider, not a CIO (Chief Innovation Officer) and a CAIO a.k.a. Chief AI Officer (yes, you read that right, it’s a full time role now!). The governance load is policies and training that don't yet exist, not ethical walls across a thousand lawyers. The buyer is a managing partner who's also going to use the thing herself, not a procurement committee three steps removed from the work.

This is the equation Lucio is built against. Multi-model, so you aren't paying a platform markup over commodity intelligence. Integrations shaped to the DMS and billing systems SMB firms actually run on, not the iManage-only stack a thirty-jurisdiction firm needs. Policy, playbook, and training scaffolding included rather than assumed, because the data says you don't have it yet and won't build it on your own. Pricing that doesn't require a twenty-seat minimum to start. And the part I care about most, designed to be used by the partner, not demoed at the partner. Vasu, our co-founder, has a line about this that's blunter than anything I'd write: most legal tech doesn't fail in demos, it fails at desks. The product that lives at desks in a sub-200 firm is a different product from the one that lives in an AmLaw innovation lab.

I'm not going to pretend that's neutral analysis. It's the bet we're making. The post-LinkedIn debate about Mike is the clearest articulation I've seen of why we made it. If your firm is sub-200 and you've been wondering whether the Harvey or Legora pitch you sat through last month was actually built for the firm you run, the question Will Chen put on GitHub is the one worth asking out loud at the next renewal meeting.

The autumn renewals are going to be interesting. Mike made the bill itemised. The firms that win the next eighteen months are the ones who use the itemised bill to ask better questions, not just cheaper ones.

If you'd like to see what those better questions look like for a firm your size, that's a conversation I'd genuinely like to have.