Your Agent May Not Really Be An Agent – Artificial Lawyer

In the latest episode of Law Punx we hear from Jake Jones, co-founder of Flank, about how your AI agent may not really be an agent. In the 8-minute blast we explore what agents really are, how they work, and why lawyers tend to misunderstand their role.

An absolutely critical part of this is not just around autonomy, but the aspect of giving an agent ‘an entire scope of work’. And this is key, as AL has explored before, lawyers tend to want to think of AI and related tools as ‘assistants’ that operate at the fringes of their needs, rather than as ‘automators’ that handle a whole stream of work. Of course, there are humans in the loop, but the goal is to say: that section of work, that’s going to be automated.

But, let’s hear from Jake. Press Play to watch inside the page.

–

This episode and four other Law Punx podcasts are on Spotify, which also features Electra Japonas, Horace Wu, Todd Smithline, and Richard Mabey.

The Spotify current episode is here – on agents.

The Spotify Law Punx site with all episodes is here – where you can listen/watch all of the episodes in one place.

And the same, but just an audio version, is available on Apple Podcasts.

AL will publish a new episode each week and there are already more that have been recorded and are in the queue to go live. If you’ve got something you’d really like to say on Law Punx, then drop AL a line.

—

Key Takeaways:

People do not understand what an agent is.

Agentic AI can autonomously make decisions and take actions.

Many tools are misclassified as agentic AI.

True agentic AI can understand user intent and adjust plans accordingly.

Linear orchestrations do not involve autonomous decision-making.

Human oversight is still necessary in agentic AI systems.

Deep domain awareness in legal tech is invaluable.

Building a new paradigm requires both insiders and outsiders.

AI tools should interface directly with non-experts.

The future of legal tech lies in fully autonomous systems.

AI Transcript

Richard Tromans here again at Law Punx. And today we’ve got a great guest. We’ve got Jake from Flank and he’s got a very strong point to make about agents. further ado, Jake, take it away.

Jake (00:41.272)

Thank you, Richard. So my point is that most people do not understand what an agent is and therefore many people are misusing the term and misusing the technology of agentic AI. So what is agentic AI? Agentic AI is something, an AI system that can autonomously make decisions and take action.

in the real world. So this, an example of this might be that I have an agentic AI system that I ask to book a hotel room for me. It understands my desires. It calls up the hotel. One hotel says I’m busy. It tries to negotiate with that hotel and say, about a different room? It knows my financial limits. It knows my preferences for the, the, floor the room is on, et cetera.

it can book that hotel or it can go to another hotel and book that hotel. It can make the payment. It can do everything basically. And it can come back to me and say, job done. So it operates as if I have a personal assistant and I say to this personal assistant, go and do it. And it can make decisions. It can plan. It can recursively fix its plan. It can fail. It can debug. It can do all of that kind of stuff. It is not just a button you push that gives you an output.

And too often, agentic AI is, the term agentic AI is being applied to software, especially in legal, where you press a button and you say something like, here’s a document, review this document for me and tell me if it’s compliant. And it comes back and it gives you a result and it’s redlined and everything else, but it’s just a linear orchestration. It’s just been executing one API call after another API call after another API call. And it’s going to do those API calls.

in spite of what the outputs at each stage are and in spite of how high quality the output is, because it’s not self-reflective, it’s not aware of what it’s doing, what it’s not doing, how successful or not each step is and everything else. So what we’re seeing now is a whole wave of products for flooding the market, a whole wave of press releases flooding the market saying, there’s this new agentic AI tool on the market or agents for law and

Jake (03:08.606)

It makes it incredibly difficult for anyone buying agentic AI or any AI or any software for legal to understand what an agent is. Because the term agent is being given to software that is just an AI product that does one thing at the press of a button that is used by the expert. An example.

that I gave before was document review, but you can get actually agentic tools. The term agentic can be applied to AI tools that do many steps still. And this is still not agentic. So an example would be there are these products that say we are, you can build your orchestration, you can build your workflows, and therefore it must be agentic.

So it might initially receive the email and then it from that email it will triage it to another system and then that system will review it and then another system will do the red lines, whatever. It’s an orchestration that has five or 10 or 50 steps. But again, in these kinds of linear orchestrations that are non-agentic, no decisions are being made. There’s no…

autonomous decision-making, no intelligent decision-making. There is no AI looking at the process that it’s currently executing, checking it’s working and saying, am I fulfilling the initial intent of the requester? Instead, it’s a button being pressed and a process being executed. And it might be a super complex process being executed.

but it’s still a process being executed. So it takes us back to the decision tree days. And it actually, we saw this with the two of the massive legal tech names releasing their agentic orchestration tools. And immediately some of the old school decision tree automation guys came on on LinkedIn and commented saying, I’ve seen this before five or 10 or 15 years ago. So.

Jake (05:23.852)

That’s what agentic AI is not and what agentic AI is and what it’s capable of is receiving a request from a business user, from a non-expert, from a non-lawyer. It’s capable of receiving a request that’s entirely unexpected and that there’s no specific orchestration that’s built for this one specific kind of request. The request could be anything. And an agentic system can

take that request can understand what the user intends and based upon everything it has access to. So documentation, tooling, MCP servers, external systems, can from all of that create a plan, begin to execute the plan and at every single step check it’s working and adjust the plan and even stop, go back to the requester.

and check that what it’s doing is the right thing. Basically, it begins to act exactly as a member of your team would act, exactly as an expert would act.

Richard Tromans (06:35.859)

Yeah, absolutely. And I personally, just to interject briefly, I’ve seen this all over the place. I mean, basically, they’re no different to a toaster or a washing machine. As you say, there’s no sort of intelligent independence going on there at all. But you made another point just off camera, which I want to touch on because I think it’s part of this argument, which is that people misunderstand the

purpose of agents, you were talking about real agents, like you were just talking about. Just tell us about that. What do people don’t understand in terms of lawyers and agents?

Jake (07:10.04)

So my belief is that agentic AI enables you to entirely replace a human expert in a given scope of work. That doesn’t mean you don’t have supervision. It doesn’t mean there’s not a human in the loop, but it does mean that you hand an entire scope of work, not just a process, not just a task, not just a workflow, an entire scope of work to an AI system.

that is agentic and it can interact directly with non-experts. It doesn’t need to be used like traditional software and like a lot of the agentic, fake agentic tools that we see today. Instead, it proactively interfaces with the people who need that expertise. And all we’re seeing flood the market at the moment are AI tools used by lawyers

instead of fully autonomous systems that have legal expertise and can resolve legal requests within a given scope.

Richard Tromans (08:18.271)

Yeah, absolutely. And I’ve alluded to this in a kind of connected article, which is basically it’s the difference between AI that behaves as an assistant on the edges of that lawyer’s capability and AI that acts as an actual automator for, know, does that entire stream. And it’s interesting, it? I mean, just to wrap up, just last point, why do you think that is? you know, lawyers are smart people, legal type people are smart people.

does this, know, they don’t seem to get it.

Jake (08:51.926)

I think one of the biggest aspects to this is that most of the people who would found a legal tech company have deep experience in operating as a lawyer within the existing paradigm of how law is done. So they’re solving many of the problems that they have encountered themselves as lawyers in legal practice or in-house. And in order to perceive an entirely new paradigm,

Sometimes it takes an outsider who isn’t building from the inside out, but is building from the outside in. Exactly. Yeah. That said, I really want to emphasize the fact that this deep domain awareness and experience is invaluable. This is not a case of saying only non-lawyers can build a new paradigm for law at all, or that non-lawyers should be building a new paradigm for law.

Richard Tromans (09:53.343)

Yeah, absolutely, absolutely. Well, to be continued. Lots to unpack there, but some great central points. Thanks, Jay. Got to leave it there.

Jake (10:01.688)

Thank you.

—

Legal Innovators Conferences in London and New York – November ’25

If you’d like to stay ahead of the legal AI curve then come along to Legal Innovators New York, Nov 19 + 20 and also, Legal Innovators UK – Nov 4 + 5 + 6, where the brightest minds will be sharing their insights on where we are now and where we are heading. 

Legal Innovators UK arrives first, with: Law Firm Day on Nov 4th, then Inhouse Day, on the 5th, and then our new Litigation Day on the 6th. 

Both events, as always, are organised by the awesome Cosmonauts team! 

Please get in contact with them if you’d like to take part.