Legal AI will create more work, but not for lawyers

Every wave of generative AI produces the same two headlines: “AI is destroying the industry” and “actually, AI is creating more work for the industry.” Whether this is for graphic designers, developers, or copywriters, the debate plays out the same way across sectors. Legal is no exception.

Against this backdrop, the legal industry provides a useful test case as it’s one of the first professional domains where specialised, workflow-specific AI has matured enough to show real usage data, not just speculation. The data suggests that the legal sector’s use of AI doesn’t map neatly onto either of the typical headlines.

To start with the obvious point, cheaper, faster legal AI tooling doesn’t just replace existing legal work, it unlocks new legal work that wouldn’t have been achieved before.

This is classic Jevons Paradox: when a resource becomes more efficient to use, consumption of it goes up, not down. More contracts get reviewed, more clauses get flagged, more compliance questions get asked, simply because the costs associated with asking have collapsed.

This is where Box CEO Aaron Levie is completely correct in his recent case for why AI means more legal work, not less. Where I think the argument breaks down is in his conclusion about who will fulfil that new demand.

His argument is essentially historical, as legal headcount has grown alongside every previous wave of workplace technology.

He references American Bar Association figures showing US attorney numbers climbing from roughly 400,000 in 1975 to over 1.37 million by the end of 2025, and argues AI extends that same curve because courts remain a bottleneck, and more AI-generated disputes and filings will require more lawyers to adjudicate them.

That’s a reasonable extrapolation if you assume the demand created by AI will be fulfilled using the same pipelines and frameworks as before. For example, generalist AI tools, such as Claude and Copilot, feeding an unchanged system of law firms and courts.

But that’s not what I’m seeing play out. The demand isn’t hitting a static system, instead it’s hitting a legal market that’s simultaneously being restructured by specialised, workflow-native tools built specifically for legal tasks like contract review, due diligence, and compliance checks. These tools don’t just add capacity to the existing pipeline, they change the pipeline itself, and who owns it.

Companies don’t have to outsource legal work any more

Legal has, for decades, been outsourced as a default. Large corporations paid large law firms premium hourly rates for work that was often mechanical and replicable.

Large law firms have traditionally relied on graduates to scan documents, review cases, and compile summaries and arguments. That work, which once took days, can now be done in minutes, perhaps seconds!

It follows that because AI can produce legal work so much faster and cheaper, law firms are being forced to reconsider the hourly billing model. As AI tools replicate and automate specific legal workflows, companies will be able to handle more legal work in-house, reducing reliance on outside counsel.

This a structural shift away from law firms, not toward more of them, and it’s the same shift we’ve watched happen in every domain where specialised AI tooling matures faster than the incumbent service can adapt.

Where the new demand actually lands: in-house teams

Levie’s framing focuses on the courtroom. The courts are a genuine bottleneck, and one AI hasn’t touched. However, most company legal work will not reach a courtroom, as it is work such as contract review, policy checks, claims processing, and compliance monitoring. These are high-volume, rules-based, and repetitive tasks that have been historically expensive.

Take insurance claims handling. The process follows a standard pattern of reviewing a policy, checking coverage against the terms of a claim, building a timeline, flagging missing documentation, and drafting a request for more information.

The courts are nowhere to be seen. Instead, it requires careful, structured judgment applied at volume, which is exactly the profile of task specialised AI is best at accelerating, with a human still in the loop for sign-off.

That’s where I’d expect the “expansion of the legal market” to actually show up: not as new billable hours at law firms, but as new capability inside in-house teams and business functions that were previously priced out of doing this work themselves.

When an industry begins to debate whether AI will “create more jobs” or “destroy jobs”, both sides are often arguing about the increased volume of output, while missing the more important question: where does the volume land?

Generalist AI tends to expand demand within existing structures. Specialised, workflow-native AI tends to relocate that demand to whoever is closest to the workflow. In legal’s case, that’s not the courts, and it’s not law firms billing by the hour. It’s the in-house teams and business users who, for the first time, have tools built specifically for the job.