The Brief, the Bot, and the Fine Print

Give an AI tool a hundred emails and ask it when a deal fell apart. In seconds, it might find the first mention of a missed deadline. It might also skip the email where both sides agreed to extend it. Either way, you'll get a neat timeline.

That's why the arguments at either extreme leave me cold. The enthusiasts are too quick to call that timeline finished. The skeptics are too quick to dismiss what it took to get there. Anyone who has spent an afternoon reconstructing a deal from six overlapping email threads can see the value in a head start. Anyone who has had to explain the deal to a client can see the danger in relying on the wrong date.

The good

Some of the best uses of AI in law are wonderfully ordinary. Pull the dates from these emails. Find the provisions that mention renewal. Turn this jumble of meeting notes into a list of open questions. I like these jobs because there's a source I can go back to and a clear point at which the work is done.

There's an access-to-justice argument here, too. A client with a modest dispute can burn through a good chunk of their budget paying someone to figure out what happened. If a firm can get through that first pass faster, verify the result, and pass along the savings, the client may have more money left for the work they came in for.

The bad

In legal work, the exception is often the whole story. A summary can correctly say a contract requires thirty days' notice and omit the amendment that gives this particular client sixty. Every sentence in the summary can sound reasonable. The problem is the sentence that never made it in.

And somebody has to find it. If you spend forty minutes checking a draft you could have written in thirty, the tool gave you a new job and called it a time savings. I'd judge an AI workflow by what it takes to get to an answer I'd put my name on.

The ugly

The most consequential part of a prompt may be what you paste underneath it. Say you ask an AI tool to tighten an email about a client's settlement strategy. You get a better email. What you may never see in the chat window is whether the provider keeps the original, whether someone can access it, or whether the interaction can be sampled for human review. Those answers can depend on the product, the account settings, and the agreement your firm actually signed. A promise about model training still leaves you with questions about retention and access.

For lawyers, that's where confidentiality and privilege enter the conversation. Confidential client information extends well beyond privileged communications. If the material includes legal advice or litigation strategy, there's also a question about how sharing it with the provider could affect attorney-client privilege or work-product protection. It's considerably easier to have that conversation while the email is still in your drafts.

Where Delphoria comes in

I built Delphoria Learning to teach this in a way people can use on a Tuesday afternoon. Give a team a fictional matter, an AI tool, and a job to do. Have them check the account settings and the provider's terms. Could someone review their prompt? What happens to the file after the chat ends? Then have them trace the tool's answer back to the documents and see what it missed. Those are skills people can carry from one product to the next.

Through Delphoria Legal Intelligence, we help firms make those same decisions at the workflow and infrastructure level, including when to keep sensitive matter data in an environment the client controls. In eDiscovery, that can mean building a review process that lets people sample the documents AI ranks lowest. You want to know whether the tool found the important emails. You also want a way to find out what it buried.

I want lawyers to get that afternoon back. I also want us to know exactly what we handed over to get it.

Learn more at delphoria.ai

Previous
Previous

The Prompt Was Cheap. The Follow-Up Was Expensive.

Next
Next

How Two Routers Can Disrupt an Entire Aviation Network—and What Law Firms Can Learn From It