In the legal profession, AI upgrades are coming straight for the bottom line.
With tech shrinking hours of administrative work into mere minutes, the long-held tradition of “billable hours” no longer measures how much a lawyer does for a client. To keep up with the times, some Pittsburgh firms are starting to charge for the software they’re using, rather than the time tasks used to take.
At the legal tech startup Squary AI, cofounder Justin Waltrip is seeing this firsthand. His company uses large language models (LLMs) to help mid-market law firms — especially ones that are in document-heavy practices like personal injury or employment law — automate some of their most time-consuming work. That can look like generating medical history timelines or drafting deposition questions in a fraction of the time it would normally take.
“At this stage, I think most lawyers recognize that these tools aren’t going away, and it’s undeniable that there is a lot of power in some of what it’s able to do,” Waltrip told Technical.ly. “We’re not saying that it’s here to replace anyone’s job, but there are certain things, like someone spending 20 hours of their time reviewing every single word of every single file, that [aren’t] necessarily the best way to handle things.”
The time savings could mean lawyers make less money, even if they’re bringing the same expertise to the table. Some firms that Squary works with are shifting to a flat fee for cases or passing the cost of Squary’s services directly on to clients, according to Waltrip.
Like in other fields, paperwork is where AI tools are creating the most dramatic difference for lawyers, impacting the two hours that a 2025 Bloomberg Law survey found lawyers spend each day on administrative tasks.
That’s what Curtis Wadsworth of the one-man law firm Nerd Lawyer is doing. Wadsworth uses models like Anthropic’s Claude to shave down time on administrative or backend work. As a result, he bills via project-based fixed fees.
Client intake, for example, used to take around three hours, he said. Now it’s 45 seconds.
“I don’t need a paralegal putting this thing together,” Wadsworth said, “so, I mean, it makes it cheaper for everybody and more efficient.”
Squary noticed this shift, too, which is why it spun out of Cognistx, a Pittsburgh-based applied AI company that emerged from Carnegie Mellon University back in 2015.
The firm was creating general Q&A systems using more traditional machine learning techniques last year, Waltrip said, building customer-facing chatbots for the Pittsburgh International Airport or Duquesne Light, for example.
When the company began getting significant traction with its legal clients, Waltrip and his cofounders Sanjay Chopra and Jagriti Pandey decided to create a new venture unattached to Cognistx’s legacy investors and contracts, he said.
Squary’s still waiting on formal paperwork to make it official, but it’s already raising a seed round to build out a sales and marketing team.
Using AI in a space that can have such a significant impact on people’s lives or businesses, though, raises the obvious concern: What about hallucinations?
Tackling AI’s trust problem
Lawyers have been caught citing fake legal cases in court filings across the country.
While hallucinations are always a possibility when working with LLMs, Waltrip said, there are a few techniques Squary uses to reduce them.
Hallucinations are more likely when input data is unclear. For example, a scanned PDF may be of bad quality and hard to read. Filling those information gaps before running a query is essential, according to Waltrip.
“We spend a lot of time making sure that whatever source data that we’re dealing with is represented in a way that the model is going to be able to understand,” Waltrip said.
Then Squary’s team picks which models are best for the task and engineers all the prompts. Outputs are treated as drafts and include inline citations, too, according to Waltrip.
“We have full control over the end-to-end process,” Waltrip said, “and so if something does go wrong — or not even wrong but just not the exact way that [the lawyers] would do it — we actually have the ability to go in and change the behavior of the system right there.”
The issues that come up more often than hallucinations, according to Wadsworth, are faulty reasoning. The AI might cite a real legal case but not accurately interpret what that case means in context.
Human expertise is still deeply important.
“It’s like working with the senior associate that knows what they’re doing enough but isn’t quite there yet,” Wadsworth said.
As for overall workload, the tech upgrades generally aren’t changing much in Wadsworth’s experience.
Burnout is still common in the legal profession, according to the Bloomberg survey. On average, attorneys felt burnout in their jobs 42% of the time, with mid-to-senior-level associates experiencing the most burnout at 51%.
The real hope, according to Wadsworth, who’s been adopting AI tools in his legal work since 2017, is in how this tech could speed up the entire judicial process.
“I’m hoping at some point it works more towards judicial efficiency too,” Wadsworth said, “so we don’t have these cases that are dragging out for years and years, although that’s a tactic for most big companies.”
