How YC startups use AI for legal

Legal is where AI moved from searching case law to doing the lawyer's work. The clearest proof is Y Combinator's own portfolio: more than a decade of startups, from 2015 to today, building the AI-native version of how contracts get reviewed, research gets done, and counsel gets delivered.

Read in order, these companies trace one shift: from contract review and redlining, to a co-pilot inside the lawyer's documents, to whole law firms run on agents. Company names and batches are public on Y Combinator (see Sources).

The shift: from faster research to the AI-native firm

The old legal model bills for time. Associates read contracts line by line, run research, draft from precedent, and the client pays for every hour. AI was first sold as a faster law library: better search, quicker citations, the same billable workflow underneath. That part is already common, and it barely changed who does the work.

The AI-native version changes the work itself. An agent reads the whole data room, redlines the contract against your playbook, drafts the patent claim, and prepares the litigation timeline, while the lawyer moves to judgment and the client relationship. The newest wave goes further: it is not a tool sold to firms, it is the firm, an AI-native practice that turns legal work around in hours instead of weeks. The companies below are split across both fronts.

The companies, oldest to newest

What they have in common

• They do the work, they do not just find it. The unit of value moved from a search result the lawyer still has to act on to a redline, a draft, or a diligence memo the agent produces.
• They run on your own precedent, not generic templates. The strongest ones redline and draft against the firm's past language and the client's playbook, because in legal the standard is your standard, not the internet's.
• They are narrow on purpose: contracts, patents, M&A diligence, litigation, corporate work. A focused legal agent beats a general chatbot, because the cost of a wrong answer is a real liability.
• The newest wave sells the outcome, not the software. The 2025 and 2026 batches are AI-native law firms: you send a matter and get finished legal work back, with the AI inside instead of for sale.

How to copy this in your startup

1. Start with the legal work you already repeat: NDA and vendor contract review, not your bet-the-company litigation. Point an agent at the high-volume, low-judgment documents first. Module 4 (Internal Tools) frames this as automating the repetitive internal work before the high-stakes work.
2. Give the AI your playbook, not just the contract. Feed it your standard positions and past redlines so it flags deviations the way your lawyer would, instead of generic risks that do not apply to you.
3. Keep a lawyer on the final call. Let AI do the first read, the first redline, and the diligence pass; let a human approve anything that gets signed, filed, or sent to the other side.
4. Use AI to make outside counsel cheaper, not to skip it. Have an agent prepare and summarize so your lawyer spends billable hours on judgment, not on reading every page from scratch.

Running legal work this way, with an agent that reviews and drafts while a lawyer owns judgment, is exactly Module 4 (Internal Tools) of AI Operating System for Startups.