AI for legal teams: what works and what doesn’t

AI is disrupting legal at a dramatic pace. But how many legal professionals actually know how to unlock the power of AI? 

Jake Jones, Chief Product Officer and Co-Founder at Flank, shares his insight on the future of legal AI and some practical tips on what legal professionals should look out for in 2025. Flank’s mission is to develop AI agents that resolve requests from the business. Jake is on the frontlines of the product team doing just that. 

—

About six months ago, at a conference, the hosts asked a room full of lawyers to raise their hands if they were using AI in their day-to-day work. Just three percent of people put their hands up. 

Usually, when we speak to those people who aren’t using generative AI in their day-to-day work, more often than not, they're astonished at what's possible. Because they looked at something six months ago or even a year ago, so maybe they didn't appreciate the rate of change in that time. But the development is off the charts. 

Unprecedented pace of change

For example, OpenAI’s o1 model was released a little more than a month ago and it’s mind-blowing (though the gap between o1 and previous models depends on how well you prompt). Putting prompt engineering to one side, the gap between o1 and the first version of, say, GPT-4o, which came out only a few months ago in May, is astonishing.

With o1, you can do document reviews of relatively complex contracts. And you can do that review with minimal prompting, whereas with GPT-4o, review requires a huge amount of prompt engineering. 

But understanding exactly what is already possible with current technology, and trying to stay on top of it, is no easy task. The more hands-on you can get with this kind of technology, the better the representation you’ll have of what AI can do. 

But the more hands on you get with this kind of technology the better the representation you’ll have of what AI can do.

The AI companies building on the application layer understand AI better than anyone because they have to make it do stuff in the real world. 

Planning for the legal AI landscape of 2025

If I were a lawyer looking at the year ahead, I’d start by finding the problems that are already entirely solvable by AI.

What can I automate right now? And that includes fairly basic stuff like automating away templating. I actually think in 2024 there shouldn’t be any kind of document that isn't automated.

I would also automate the high-volume work that is taxing but not complex. If you have high-volume queries that can be handled now by AI, then you’re in a position to automate it away entirely.

What I wouldn’t do, which is as important as what I would do, is to try and use AI to help me tackle some of the more complex work, because AI is going to get progressively better over time. 

So the approach I’d take is to almost have a roadmap of the pain points that I have, the work that I have on a daily basis, the kinds of tickets that land on my desk on a daily basis. And I'll grade them from lowest complexity to highest.

That's my roadmap, I start with the lowest and see where AI today gets me. As soon as I reach a point where the AI starts to falter a little bit and I lose confidence in the AI, I stop. 

That’s when I wait for the next foundation model, or I wait for the next Flank or Juro to come along and solve that in a way that I can trust.

‍Listen to the full episode of Brief Encounters to discover how Jake and the team at Flank are solving legal bottlenecks with AI Agents.