Legal AI is having a moment.
The market has seen a surge in new startups, high-profile funding rounds, large LLM providers launching legal specific solutions, and bold claims about the future of legal work. From contract review tools to precedent search engines, the message is clear: AI is here to help lawyers work faster and smarter.
And yet, behind the hype, a quiet reality is setting in for many legal teams. These tools promise a lot—but are they really solving the problems lawyers care most about?
What's actually under the hood
Layering legal interfaces on top of large language models (LLM) like OpenAI's GPT or Anthropic's Claude has become the default approach across the industry—startups and incumbents alike, including Carta. It's a fast, practical way to bring capable AI to legal work. But the foundation is no longer the differentiator. What a provider builds on top of that foundation is.
These models are general-purpose by design. They have been trained on massive amounts of public data—from Wikipedia to Reddit to open-source legal documents. That breadth gives them impressive fluency, but it does not give them depth in any one firm's approach, any jurisdictional specifics, or any client's preferences.
So when AI tools built on these models tackle legal tasks, they are often doing it from the outside in. They can spot patterns in language, but not always in context. They can generate summaries and suggestions, but not necessarily ones that reflect your standards, risk profile, or terms.
The question worth asking: Does the tool built on top of your model actually understand your firm?
Standard problems, standard results
The logic behind many of these tools ( regardless of what model sits underneath) is to identify common legal problems and build standard, one-size-fits-all solutions. Contract review is the classic use case. If every non-disclosure agreement (NDA) follows a similar format, the thinking goes, why not use AI to mark up the red flags?
The problem is that legal work is about knowing why something matters, how your client feels about it, the jurisdictional or counterparty nuances, and what the commercial context demands.
What looks like a red flag in one situation might be entirely acceptable in another.
The result? AI that technically works, but practically underwhelms. It generates outputs lawyers still have to review, second-guess, and often rewrite. The risk of AI-generated errors or missed context means the tools can create a new layer of work—not less.
Why this isn't enough
When lawyers are asked what they want from AI, the answer is rarely a clever draft. It is a faster path to the right answer. That means tools that know how their firm works. Tools that understand the difference between "acceptable", "preferred," or "non-negotiable." Tools that surface relevant client precedent without anyone needing to search for it.
This is where the current generation of legal AI falls short. It is not that these tools are poorly built. It is that they are not designed for the specific ways law is practiced inside individual firms. And they require lawyers to do more—not less—to make them usable. That gap has nothing to do with which LLM powers the tool. It has everything to do with how much firm-specific context, precedent, and workflow the tool is built to absorb.
The adoption paradox
Many legal tech pilots fail to move past a few enthusiastic early users. This is something we are seeing become public more and more.
Lawyers are busy. They will not adopt tools that force them to change how they work, switch between interfaces, or second-guess the results.
And yet, this is exactly what most AI tools ask them to do. A tool might be sophisticated, but if it requires a new workflow—or if the outputs still need checking—it is just one more item on the to-do list.
The paradox is that the more powerful the tool, the more consequential its errors. Lawyers can end up spending more time reviewing AI outputs than they would have spent completing the task themselves. When this knowledge sits within the brain of a partner, distilling it and training the AI becomes a full time job—one that might have to be repeated when the open-source model updates or the underlying code changes.
Getting to the right model
Another challenge with many legal AI tools is that they address only one isolated part of the workflow: reviewing a clause, suggesting a redline, or generating a draft.
The problem is that legal work does not happen in isolation. Every review, every markup, every negotiation, every compliance check sits within a broader context.
If the AI does not understand that bigger picture, it cannot make the right decisions at each step.
That is why solving just one piece of the workflow is not enough.
What works instead is a model where AI is built into the legal workflow—not layered on top of it. When AI is built around the entire legal process rather than a single task, you do not just get better individual tools. You get faster, safer, more predictable outcomes—the kind lawyers and clients actually want.
That is the difference. Legal AI should remove steps, not add them.
DISCLOSURE: This publication contains general information only and neither eShares, Inc. dba Carta, Inc. (“Carta”) nor Carta Law is, by means of this publication, rendering accounting, business, financial, investment, legal, tax, or other professional advice or services. This publication does not give rise to any lawyer-client relationship, is not a substitute for such professional advice or services and nor should it be used as a basis for any decision or action that may affect your business or interests. Before making any decision or taking any action that may affect your business or interests, you should consult a qualified professional advisor. Carta does not assume any liability for reliance on the information provided herein. © 2026 eShares, Inc. dba Carta, Inc. All rights reserved. Reproduction prohibited.
