The Grey Area Between What AI Can Do in Legal and What Still Needs a Lawyer

It’s Tuesday afternoon. A business unit lead has just messaged you that they've found an AI tool that drafted their vendor contract. It looks good to them so it just needs a quick check from legal before it goes out.

You open it and within the first few clauses, you already know this isn't going to be a quick check… it never is. 

This is the reality that no legal AI demo has quite captured. Legal teams are being asked to move faster, cover more ground, and do it all with headcount that hasn't kept pace with demand. AI was supposed to ease that pressure. 

For many teams, it has (in parts). But it has also introduced something nobody put in the brochure: a new kind of uncertainty about where the technology's judgment ends and yours needs to begin.

That line is blurry.

The Promise We Were All Sold

The optimism about AI in legal is not unfounded. It can draft, summarize, extract, compare, and field routine queries at a scale no team of lawyers could match. For stretched legal functions fielding hundreds of requests a quarter, that is not a small thing.

And the early wins were real. Contracts turned around faster. Standard queries answered without consuming billable hours. Junior team members unblocked from the administrative drag that used to swallow their days. There are legal teams who will tell you, honestly, that AI changed the way they work for the better.

But between the promise and the practice, a quieter problem emerged. One that doesn't show up in efficiency metrics or vendor case studies. It's the persistent feeling that despite everything, the volume hasn't really dropped. Requests keep coming and the genuinely complex work still lands on the same desks it always did, except now it arrives alongside a draft that someone else's AI already had a go at.

So, while the tools got smarter, the grey area between what AI can do and what still needs a lawyer got wider.

The Problem: The Line Keeps Moving

In-house legal teams have always operated in the space between "this needs a lawyer" and "this probably doesn't." What's changed with the introduction of AI is that these intelligent tools have made that space both larger and more consequential.

Context Dependency

A standard NDA clause isn't standard when the counterparty is a competitor, or when the jurisdiction shifts, or when the business relationship carries history that isn't captured anywhere in the document. AI reads the clause, but it doesn't read the room. And the gap between those two things is precisely where legal judgment lives.

Escalation Problem

Most teams don't have a consistent, defensible answer to the question of: “when does a routine request stop being routine?” It tends to live in the instincts of whoever happens to pick it up, meaning the threshold shifts depending on the day, the workload, and who's in the office. Unfortunately, that inconsistency opens the door to greater risk.

Liability Question

The liability question tends to get glossed over in AI enthusiasm. When an AI-assisted output turns out to be wrong, who owns that? The lawyer who reviewed it? The team that deployed the tool? The answer is rarely clear, which means the cautious move is to treat every AI output as if it needs the same scrutiny as something written from scratch. And if that's the approach, the efficiency gain starts to look a lot thinner.

Trust Deficit

Underneath all of it is a trust deficit that runs in both directions. Business stakeholders don't always know what warrants legal input, so they either send everything — or worse — send nothing until it's too late. Legal teams, meanwhile, don't always have a clear mechanism for signaling where the real risk sits. This results in a function that is simultaneously overloaded and underleveraged.

The grey area, in other words, is a judgment and infrastructure problem that the AI boom has simply made harder to ignore.

What the Grey Area Actually Costs

Imagine there’s a business team that used an AI tool to generate a supplier contract, felt confident in the output, and sent it off for signature before legal saw it. They'd used the tool before and it had always been fine. Legal only heard about it when a dispute emerged three months later over a limitation of liability clause the AI had drafted in a way that made perfect syntactic sense and almost no commercial one.

There's the GC who, twelve months into her organization's AI rollout, realizes she can't actually answer the question of whether her team is more efficient or just busier. The requests are moving faster. Whether the right ones are getting proper attention is a different question, and she doesn't have clean data on it.

Then there's the CLO who gets asked in a board meeting whether the company's use of AI in legal is governed, and who pauses, just for a beat, before answering. Not because it isn't, but because "governed" is doing a lot of work in that question, and the honest answer is more complicated than the room has time for.

And lastly there's the in-house lawyer who reviews every AI-assisted output personally before it goes anywhere, because the cost of missing something feels higher than the cost of the extra hour. Which means the efficiency gain the tool was supposed to deliver has been quietly redirected into a new form of the same overhead.

These are the ordinary, invisible costs of operating without a clear boundary.

Where Does AI End and the Lawyer Begin?

Most legal teams are still figuring this out because the question itself has been framed too narrowly. The conversation tends to start and end with capability. Can AI draft this? Can it review that? Can it handle the volume?

The better question is: “Does this actually need a lawyer?”

Because when you look at what genuinely floods a legal team's inbox on any given week, such as routine vendor queries, standard NDAs, internal policy questions, and low-risk contract requests, a significant portion of it doesn't require legal judgment. It requires legal infrastructure. A consistent, reliable way of handling predictable work so that it moves without consuming the attention of someone whose judgment is genuinely scarce and genuinely valuable.

That's where AI belongs. As the operational layer that handles the work that was never really about legal thinking in the first place. The drafting that follows a pattern. The triage that applies known rules. The questions that have been answered a hundred times before and will be answered a hundred times again.

The high-stakes work is different. Nuanced negotiations, novel risk, anything where the context, the relationship, and the downstream consequences converge in ways that don't reduce to a template, still needs a lawyer. Not because AI can't produce an output, but because the cost of the wrong output is too high, and the judgment required to know the difference isn't something you can automate.

The teams getting this right have stopped treating this boundary as something to discover after the fact and started designing for it deliberately. They've built a structure where AI handles the volume and the overhead (i.e. context switching, admin drag, repetitive triage that burns hours without adding judgment), and lawyers are protected for the work that actually requires them.

That shift from "how do we use AI?" to "how do we build a legal function where AI and lawyers each do what they're actually for?" is a harder problem than it sounds. But it's the right one to be solving.
‍

Key Takeaways

The grey area between AI and the lawyer isn't going to resolve itself. If anything, as the tools get more capable and the pressure on legal teams continues to build, the stakes of not having a clear answer will only grow.

But this is a solvable problem. The teams demonstrating that have simply been deliberate about something most legal functions have left to chance: defining where AI-assisted work ends and where legal judgment begins, and building the infrastructure to make that boundary real and enforceable in practice.

That's the problem Checkbox was built to sit inside. To make sure a lawyer’s judgement gets applied to the right things. The Legal Front Door concept starts exactly here with the idea that how work enters a legal function is just as important (and arguably even more important) as how it gets done.

If that tension feels familiar, it might be worth exploring what a more deliberate boundary could look like. Schedule a call with one of our technology consultants today and see what a more deliberate boundary could look like for your legal function.