Artificial intelligence can enhance legal work, but human judgment remains essential as law firms aim to deploy the technology effectively, a group of experts said Monday.

The first day of the 2026 International Legal Technology Association Conference in Nashville featured a panel of experts discussing how to exercise caution and display confidence when prompting legal AI tools. Sierra Gaines, a lawyer working in practice efficiency and innovation at Miller Thomson LLP, moderated the panel session.

During the panel discussion, Ice Miller LLP partner Michael Ott warned about the dangers of having an AI tool agree with attorney preconceptions without evaluating the AI-generated work.

"That's a perfect storm for getting things wrong," Ott said. "You still need to check everything."

Ott discussed the perils of "vibe lawyering": It refers to lawyers delegating judgment to increasingly powerful AI tools. While it might be tempting to do this, he said, lawyers risk letting the AI make the judgment on responses, facts to contest and the tone to take.

There's also an ethical framework for avoiding vibe lawyering, Ott said, noting the American Bar Association's Formal Opinion 5.12 states lawyers must understand how generative AI works and are required to supervise AI output.

More advanced legal AI tools are capable of drafting an entire brief after a single prompt, but Ott said this is a potential pitfall for attorneys since it effectively delegates judgment to the AI tool, and it may be more difficult to spot wrong information immediately.

And as legal AI tools move into consumption-based pricing, where firms are charged for each prompt or task, regenerating a brief could be costly, Ott added.

"Firms are spending a ton of money on these tools," Ian Lopez, senior manager of technology and innovative solutions at Ballard Spahr LLP, said during the panel discussion. "It's key to show the return on investment."

Lopez said law firms can show the return on investment on AI tools by tracking usage increases over time, the number of attorneys using the tool, the number of queries run on the tool, the number of documents uploaded and the number of tokens used. Tokens are data that AI models use when generating content.

One of the takeaways from the panel discussion is that picking the right AI tools also requires a detailed strategy. Lopez said Ballard Spahr measured different potential AI tools by judging platform interfaces, security and vendor engagement during the demo process. He added that firms might need multiple AI tools to fit the needs of different practice areas.

According to Lopez, Ballard Spahr successfully rolled out AI tools through one-on-one training and finding attorneys in each practice area to encourage other colleagues to use them.

Another takeaway from the discussion is that AI adoption requires an eye on quality control. Lopez said mistakes made by AI tools, often referred to as hallucinations, are a reminder that AI is not a perfect solution. He said attorneys should display judgment and knowledge before, during and after using an AI tool.

With the potential pitfalls associated with AI, it is understandable that some attorneys might be fearful of using these tools, according to Ken Lam, the lead AI transformation project engineer at Thomson Reuters. Lam spoke during the panel discussion about psychological safety and the fear of getting wrong outputs from prompting the AI tool.

Lam said attorneys should continue experimenting with AI, but without training tools on client data.

Ott said that this experimentation is "how you learn the limits of the software."

--Editing by Lakshna Mehta. 

For a reprint of this article, please contact reprints@law360.com.