Client expectations are one of the top drivers of AI adoption in the legal industry (ALLRIZE). Clients no longer judge firms on reputation and pedigree alone. They want speed, cost predictability, and clear answers to business problems (Lambreth, LawVision). These new expectations pose problems for firms whose entire value proposition has been built on prestige.
It's also a textbook case of Clayton Christensen’s Innovator's Dilemma: successful organizations fail not because they're badly run, but because they're run too well using dated assumptions. The foundations of Big Law, the billable hour, risk aversion, and the partnership model, are exactly what now make it slow to change (Xu, LinkedIn). The large firms see it coming and are responding aggressively, with some spending hundreds of millions to build their own AI platforms (Scarcella, Reuters). Unless these tools are woven deeply enough into operations, however, a half-billion-dollar platform is just an expensive line item.
General Legal is being built fully aware of this opportunity—that’s why I wanted to spend my summer here. Because the firm is AI-native by design rather than by retrofit, it does not need to completely rebuild itself while trying to stay at the forefront of client expectations. The AI triages, summarizes, and flags risk; an attorney reviews, redlines, refines, and takes accountability for the result. I get exposure to the tools and processes disrupting the future of legal work in real time, learning and iterating alongside experienced attorneys.
What is the role of a law firm when clients have access to the same cutting-edge legal tools? A founder reviewing a vendor agreement can paste the whole thing into a chatbot and get a clause-by-clause explanation, a list of red flags, and a drafted counterproposal before her lawyer has even opened the file. For routine work, that's often good enough, and pretending otherwise will have dire consequences for legal industry players in this transformative era.
Some clients trust AI. Others go a step beyond and expect it. To explore how we can design legal technology that optimizes how humans and AI systems collaborate, I ran two experiments: I used AI (1) to draft and review a contract, and (2) to turn an executed agreement into a clean, reusable form template for future deals.
Drafting and Reviewing a Contract
The first experiment was a deliberately outlandish test case: a goods-and-services purchase agreement priced in a fictional currency and governed under the law of an invented tribunal. I ran it through General Legal’s proprietary Sentinel AI and compared the output against the markup of a seasoned attorney who reviewed the same draft independently. The results were comparable.
On raw issue-spotting, the two overlapped almost completely: both flagged the unilateral exchange rate the seller could set at will, the 300% late-payment penalty, an indemnity that made the buyer cover the seller's own fraud, contradictory IP and delivery clauses, and a “sole discretion” services obligation that rendered the seller's core promise illusory.
Notably, the AI surpassed human review by identifying how the seller could collect 50% upfront, invoke a sole-discretion clause to decline work, and cap liability at $5,000—potentially pocketing half a million-dollar deal while risking only $5,000. It surfaced and synthesized important economic asymmetries the human markup missed:
- a termination-for-convenience right the seller could exercise but the buyer could not;
- a five-day deemed-acceptance window on specialized equipment unlikely to be tested that quickly; and
- an asymmetric risk structure ($5K liability cap on a $1M deal).
Even then, I would hesitate to rely on a contract without verification by a real human attorney. Unlike the AI, an attorney went beyond recommendations and made explicit decisions about where to change the document. For example, on the ten-year worldwide non-compete, the AI gave the textbook recommendation: “overbroad, narrow or delete.” The attorney instead rewrote the entire article, retitling it from “Non-Compete” to “Exclusivity” and reworking the operative language to match. The attorney made a judgment about what the client probably wants, an exclusive supply relationship the seller will actually agree to and a court will actually enforce, and then revised the paper to achieve that. Though the AI could flag problems, the attorney was needed to strategize solutions.
Turning an Agreement into a Template for Future Deals
The second experiment, the templating task, illustrated the practitioner’s subtle understandings that AI has yet to achieve. I wanted to know whether a technology order form template should include an “Inspection and Supervision” clause by default.
The AI, recognizing the clause from similar agreements, was happy to keep it. Anyone who has used AI is aware of the bias towards gratifying the user and assumption that the work the user has provided is generally acceptable. Liz, Managing Partner at General Legal, set me straight: that clause shouldn't be a default, and should only be included if the client asks for it. She explained what the language says on paper and, unlike the AI tools, how it actually gets used, ignored, and fought over in practice.
LLMs are pattern-matching machines, always tending toward accept patterns they've seen before. As such, they will suggest corrections to discrepancies in those patterns without questioning their veracity. A good lawyer optimizes toward their client’s position and knows that the most important clause might be the one a model would never question.
Leveraging AI vs. respecting human experience draws a false opposition. The legal industry’s top players will be those who effectively allocate their resources to balance the two.
TL;DR
Just Getting Started
Claude for Legal debuted less than a month ago. General Legal is less than a year old. Even Harvey, perhaps the most established AI-focused player in legal tech, is a mere four years old. Given the nascency of the space, there are many open questions about the future of AI-powered law firms and the impact of AI on the legal industry, and I look forward to exploring them this summer.
Working With Us
If you're a founder or operator looking for legal counsel built for the speed of the AI era, General Legal offers high-quality, predictable work backed by experienced attorneys focused on your business goals. Ready to see the difference an AI-native approach makes? Sign up here to get started.
Sources
- “2025 Legal Technology and AI Adoption Report,” ALLRIZE
- Scarcella, "Law firm Kirkland to spend $500 million developing its own AI platform," Reuters
- Lambreth, "Disruptive Innovation in Law," LawVision
- Xu, "When Big Law Meets AI: An Extreme Case of the Innovator's Dilemma," LinkedIn
- “Baumol's cost disease,” EBSCO Information Services, Inc.
Frequently Asked Questions (FAQs)
What did the AI catch that the human attorney missed?
Three economic asymmetries: a termination-for-convenience right only the seller could use, a five-day deemed-acceptance window on equipment unlikely to be tested that fast, and a $5,000 liability cap on a $1M deal.
What is Sentinel?
Sentinel is General Legal's proprietary contract-review AI. It performs the first-pass review - issue-spotting, summarizing, and flagging risk - before an attorney reviews and refines.
What is a "termination for convenience" clause?
A right to end a contract at will, without cause. This is risky when only held by one side, as in the test contract, where the seller could end the contract at will but the buyer couldn't.
Why not let the AI keep a clause just because it recognizes it?
Recognition isn't judgment. The AI kept an "Inspection and Supervision" clause by default; our managing partner knew from practice it shouldn't be a default and belongs only when a client asks for it. In legal practice, an understanding of how language is used, ignored, and fought over is critical.