Agents vs Copilots: which do you need?

There’s a huge amount of talk about agents right now. But what’s the difference between an agent and a copilot in the context of legal solutions and why should you care?

I’ll leave the technical underpinnings of agentic vs copilot application layers to others, but from a business outcomes POV, the simplest way to think about the difference between the two is autonomy.

Let's investigate

Agents can perform tasks autonomously, whereas copilots can’t. Not really. So for example, if we take a common AI use-case in legal: reviewing a contract. A copilot might read your contract and recommend stuff for you to do with it. Agree with this clause interpretation, action this reminder date, flag this deviation from your standard playbook, and so on.

An agent might read your contract, know all the context in the same way a copilot does - and instead of recommending actions, it takes them. It just does it.

Stepping back for a minute: who cares?

In the example of a contract review, these are two routes to the same business outcome: a redlined document. In the first scenario you were offered suggestions and accepted them, and in the second you just got the finished document.

The first scenario has a decent amount of lawyer-in-the-loop. The second has a bit less, or potentially none at all. Imagine, for example, the business self-serving autonomously on an NDA review. In most cases with an agent, a lawyer doesn’t need to be involved. This is a much bolder and more exciting solution path.

But what about the outcomes?

With my business leader hat on, if I think about the outcomes, I’m actually much less interested in how we get there. But what matters are the speed and the quality.

Speed is easy to think about. Faster is faster.

Quality is a bit more complex. Is it accuracy? Maybe it is, for now, but new models ship every few weeks and the accuracy distinctions will eventually reach a point where it won’t matter (like the SNES vs the iPhone).

A different component of ‘quality’ is judgement. I’m talking lawyer judgement - the ability of a highly trained legal professional to make qualitative adjustments based on their experience.

This could well be a differentiator. Your ability to teach and to impose your guardrails on an agent could have a huge impact on that business outcome.

Looking further ahead, it’s not hard to imagine a scenario where you buy that judgement. Your vendor sells the contract playbook along with the product, meaning you don’t even need a lawyer to implement it.

Put guardrails in place

Let’s look at how we get contracts reviewed now. At Juro, if we get an inbound NDA we allow sales reps to use our AI Assistant to review it against a playbook. They can do that because Michael Haynes is accountable for the playbook and can jump in where there are exceptions.

That’s pretty far into the lifecycle before the lawyer comes back into the loop. Sometimes he’s not needed at all. If agents are autonomous, then the amount of the contract lifecycle up for grabs increases, as generative AI can achieve parity with junior lawyers for a big chunk of contract work (there are many studies about ChatGPT taking the Bar exam to prove it).

To actually make this work day to day, in a way that’s acceptable from a risk appetite standpoint (we’ve all got fiduciary duties after all), then you need great curated content. And this is where lawyer time shifts in an agentic world. This changes fundamentally the jobs that in-house lawyers do, but in a good way.

Great playbooks, great templates, great help center, great platform so nothing goes wrong, great implementation of whatever you’re using to make sure the backstops are in place, and accessible enough that anyone in the business can use it. These preconditions are all necessary for a lawyer-lite process to function.

The need for speed

This means that lawyers become content curators or, more provocatively, puppeteers. They set the stage and pull the strings, and autonomous agents (real or artificial) handle the lifting.

As a business leader, I like it from an efficiency point of view. But as a lawyer, this is really is breaking new ground and I would bet that for documents with more variation and risk than NDAs and the like it will be a little time before we lawyers jump fully out of the loop.

So back to the top - what is better, an agent or a copilot?

Lawyer answer? It depends. Probably for the simple documents, agents, and for the more complex documents, copilots. Mulder & Scully for NDAs, Goose for cross-border licensing agreements. At least for now. But things are moving very fast indeed.