All posts

EngineeringMay 2, 2026·9 min read

How AI changes the contract-review workflow

There's a tedious version of the AI-and-law conversation that goes "AI is going to replace lawyers." That conversation is wrong, but not in the way people usually argue.

The thing AI replaces isn't the lawyer. It's the first 80% of every contract review — the parts where the lawyer is reading clauses they've read a thousand times, classifying risk against a pattern they already know, and writing the same redline language they've written for the last seven clients.

That work isn't going away. It's getting compressed from 90 minutes per contract to two minutes, with the human staying in the loop for the parts that actually need judgment.

The pre-AI workflow

Here's what reviewing a 14-page vendor MSA looked like in 2022:

1. Triage — assistant or junior associate sorts incoming contracts by urgency. Say 30 minutes per intake.
2. First pass — associate reads the contract end-to-end, marks up obvious issues, drafts redline language. 90-120 minutes.
3. Partner review — partner reviews the associate's markup, escalates the deal-specific concerns to a phone call with the client. 30 minutes.
4. Client conversation — partner walks the founder through what they should push back on, what's standard, what's weird. 30-45 minutes.
5. Counterparty negotiation — partner sends redlines, counterparty counter-edits, back-and-forth until both sides converge.

Steps 1-3 cost the founder somewhere between $400 and $800 in billable time. The actual conversation that moved the deal forward was step 4.

The post-AI workflow

What we think the right shape is:

1. AI first pass — founder uploads contract, gets a structured review with severity-coded findings, cited sections, and proposed redline language. Two minutes.
2. Founder triage — founder reads the findings, decides which deal- specific concerns to escalate. 10 minutes.
3. Counsel review — partner gets the AI-generated findings + the founder's annotations, focuses entirely on the deal-specific concerns and the cases where the AI's standard suggestion isn't right for this contract. 20-30 minutes.
4. Counterparty negotiation — same as before.

Steps 1-3 cost the founder $29 + 10 minutes of their own time + $200-300 of counsel time. The deal-specific judgment work happens in step 3, which is exactly where the value of legal counsel actually lives.

Where the human stays in the loop

A few categories AI is genuinely bad at, and where you should still budget counsel time:

• Multi-document deals. Side letters that interact with stock purchase agreements that interact with founder employment offers. The cross-doc reasoning is fragile.
• Jurisdiction-specific enforceability. "Will this non-compete actually hold up in California / Massachusetts / Germany?" The model knows the general shape; it doesn't know the local case law you'd need to bet a deal on.
• Strategic concessions. Whether to give up an audit right in exchange for a better liability cap is a deal-strategy decision, not a clause- reading one.
• Anything you'd lose your job over. Term sheets, board resolutions, M&A docs.

We surface these as warnings in the review output: "This is a side letter to a stock purchase agreement; cross-document review recommended." It's not a hard stop, but it's a flag that the AI is operating outside its strongest territory.

The compression matters

Most founders we talk to don't choose between "send to counsel" and "use AI." They choose between "send to counsel" and "sign without reading." That's the comparison that matters. A contract review that takes two minutes happens; a contract review that takes four days and $600 doesn't, because the contract has to be signed today.

Compression of the standard work is the unlock. The work that's still hard gets done by the people who are still good at it. The work that's been boring for thirty years finally goes away.