One of the scarier emails I’ve received at work came in a few years ago, before we had any approved AI tools.

A colleague was working on a new technology and had an idea for saving some time. He was going to describe it to ChatGPT and have it help him draft the patent application. He just wanted to let me know.

He was on the free version of ChatGPT, without enterprise terms or any sort of data protection.

Never have I ever reached for my phone faster to call a colleague and ask him to please pause while we talked it through. We talked. I introduced him to patent counsel. And everything was fine.

But it could have been very much not fine. Had he gone ahead, he risked the input being treated as a public disclosure that could get in the way of the patent. And by handing it over voluntarily, he might have undercut a trade secret argument as a fallback.

I’m obviously a huge fan of AI. I work at an incubator building AI companies across multiple industries. And over the last 18 months, my team and I have saved hundreds of hours and many thousands of dollars by using AI. I also show up better prepared for last-minute meetings, catch more of those buried “legal will get you X” promises in email threads, and worry less that I have missed a key message from a director or executive who won’t wait, because AI provides regular reminders.

But it’s not all sunshine and rainbows. Sometimes AI really stinks.

Your inbox isn't the problem. Neither are your meetings. It's the hours you spend on work AI could do in minutes. 

285 copy-paste prompts: built for email, meetings, admin, and delegation. Each one has a clear role, a specific task, and a defined output. 

No setup. No prompt engineering. Just copy, paste, and get your time back. Free when you subscribe to Mindstream.

Get The Free Guide

This shows up two ways.

The first is when someone outside legal uses AI to draft a legal document. A colleague drops you a note asking for a quick review of something they put together themselves. They think it’s in good shape.

At a glance, they’re right. In the franken-contract days, you could spot the indemnity lifted from their office lease and the IP clause borrowed from their web host’s terms of use. Now everything looks like it belongs together.

Then you dig in. Terms contradict each other and provisions pull in concepts that don’t fit the deal. The whole thing runs twenty to fifty percent longer than it needs to, which you know will slow everything down the minute a counterparty sees it.

So you have a choice. Fix their draft, which takes meaningful time and a careful check against your company’s standards, or start over. If you have well-documented standard terms, or an AI tool that already knows them, either path works. But the fastest route to the best result existed before anyone opened a chatbot: a short conversation with your colleague about what they’re trying to accomplish, then your own prompt to your approved tool with the right background built in, would have gotten to an effective outcome faster.

The second time beautiful slop comes up is when you’re the one using AI, like redlining counterparty paper or taking a first pass at a new template.

The output looks decent at first. But even with excellent prompting, AI tends to do too much. Nothing it adds is wrong, exactly. It just adds so many belts and suspenders that the person wearing the pants can’t walk.

So you prompt, and reprompt, and tweak. All of which takes time.

Start with your colleagues. Most people who send you an AI draft think they’re doing you a favor, because they believe they saved you a step. Your job is to reset that expectation, kindly and more than once. Reinforce that talking to you first gets them a better document, faster and for less money, than a draft you’ll have to untangle.

Say it when they send you their AI draft. Say it again when you’re asked how legal can move faster. Once people see that a fifteen-minute call gets them a clean document sooner than their own draft would, most of them will pick up the phone first.

To be clear, I am not saying you should keep the business away from AI. A colleague who uses it to organize their thinking before they call you is handing you a useful, clear picture of what they want. The trouble starts when their draft replaces the conversation, and the draft should have come from you.

Then look at your own habits. Before you hand something to AI, ask whether this is a task where it saves you time or creates more of it. A first-pass issues list on counterparty paper or a starting draft built on your own standard terms usually saves time. A negotiated document where the value lies in what you leave out often costs you more than the initial time savings is worth.

When you do use it, tell it what you don’t want. Give it your standard positions, a length limit, and permission to flag issues rather than fix every one of them. The less room you leave for belts and suspenders, the fewer you’ll have to take off.