If you've just moved in-house, you have tools I didn't have when I started. AI can summarize a 40-page agreement or mark up an NDA against your company's standard in a few minutes. Use it.
Your internal clients have the same tools, though. The sales director can paste a contract into one as easily as you can. If your advice reads like something the tool would produce, people will stop asking you and ask the tool. What they can't get from it is a lawyer who knows the business and its history, and who will tell them what is likely to happen and what to do about it.
I've spent most of my career in-house, as a general counsel, a corporate secretary, and senior in-house counsel. I also spent four years as a litigator, trying cases as first chair. Part of my work now is testing AI models on legal tasks and grading the answers the way a senior lawyer would. The models handle the law and the documents well. Their blind spots are the parts of the job that happen away from the documents.
Meet your internal clients
At a law firm, work usually reaches an associate through a partner, already framed as a legal question. In-house, it comes as a two-line email from sales or operations, and the email often leaves out what matters most. "Can you turn this NDA today?" may mean the business is about to show a competitor its current pricing as part of a deal the CEO hasn't announced. The NDA is the easy part. Sharing current pricing with a competitor raises antitrust questions that no NDA solves, and you'll only hear about the deal by asking.
Know who your client is. Florida's Rule 4-1.13, like the rule in most states, says a lawyer employed by an organization represents the organization, acting through its duly authorized constituents. The VP who needs a signature by Friday is your best source of facts and usually wants what the company wants. When those interests split, you need to know who can make the call.
Go meet people early and in person, before anything goes wrong. Sit in on their staff meetings and walk the plant floor with the people who run it. People tell their lawyer things in a meeting that they won't put in an email. They also bring in a lawyer they trust early, while a deal can still change, instead of at the signature page.
Learn the history
Every company has a past that shapes today's answer. AI can only search what you give it, and much of that past was never written down.
In one in-house role, one business channel told me a legacy supply agreement with a large customer was terrible. The commercials were bad, and most of the other terms were nothing we would sign today. The agreement had a 24-month termination notice period, and the notice had gone out before I even joined the company. The expectation was simple: when the old agreement expired, a new one would be in place, with our other business channel as the party because they were more suited for the relationship.
The customer accepted the termination and sent its own paper for the new deal. The terms were far outside our risk appetite, and the liability exposure was many times what the old agreement allowed. I redlined it in earnest and built a negotiating plan around it. Neither I nor the AI tool I used knew the history.
It came out on a call with the sales director, the general counsel, and the VP. This customer made frequent defect claims, and the old agreement's liability cap was the one term that made the account workable. The contract everyone called terrible was protecting us.
So we changed course. We asked the customer to sign the old agreement but updated to the new business channel as the new party, keeping the cap and bringing the pricing up to date. We also had to explain that the move was meant to give them better service. They said no at first and agreed a few months later.
None of that history was in the agreement or the request that reached me. It lived with the people who had been there. Other history can change the answer just as much:
How the company handled the last similar case. Before you sign off on firing an employee, ask how the company treated others who did the same thing. If someone outside the employee's protected class kept their job after the same conduct, that comparison often sits at the center of a discrimination claim. An AI review of one personnel file will never see it.
What other customers got. A concession to one customer tends to become the next customer's opening ask, and some contracts promise a customer terms as good as anyone else gets.
Old agreements that still bind. A settlement, consent order, or side letter from years ago can limit what the company may do or say today.
What the company has already said. Positions it took with a regulator or in court are hard to walk back.
To find it, read the last few years of board minutes and the claims and litigation history, then ask the people who have been there longest. The controller and the CEO's assistant often know more than the files do. If your predecessor is still around, take them to lunch.
Learn what the business is trying to do
Learn how the company makes money before you advise on how it spends or risks it. Know the top customers and suppliers, the margins, the plan leadership gave the board this year, and what each team is measured on. A sales group chasing a quarter-end number will ask for different things than a plant judged on uptime.
Legal advice is one input into a business decision. Your job is to help the company get where it's going, with risks it understands and has chosen to take. Most "can we do this?" questions are really asking how to do it safely.
Risk tolerance also changes with the deal. A term you would refuse on a small one-off order may be worth accepting to land the company's first major customer in a new market. AI doesn't know which deal is which, or which risks leadership has already accepted. So don't forward the tool's list of twenty issues and let the business sort it out. That hands the judgment to someone with less legal training and teaches them that legal slows deals down. Pick the few points that matter for this deal and say why.
Give advice someone can act on
The business needs to know what you recommend, what will probably happen, what it will cost, how long it will take, and what could go wrong. "There is a risk that" tells them none of that.
AI defaults to balance. It lays out each side and leaves the choice open, which looks like careful legal writing, but a CEO deciding whether to cut off a customer needs an answer. Florida's Rule 4-2.1 lets a lawyer refer to moral, economic, social, and political factors as well as the law, and its comment warns that purely technical legal advice can sometimes be inadequate.
Length matters too. In one test. of an AI model, a business executive asked for a lawsuit against a customer and wrote that he didn't need a memo. In the last round, all four answers ran between about 1,600 and 3,000 words. The two that rightly advised against filing for now read like briefs, dense with commercial code citations. The right advice was in there, but the executive would have had to dig for it. Put your recommendation in the first two lines and the support below it. Remember your recipient is likely going to read it on a cell phone. This needs to be clear, concise, and to the point. AI does not draft that way.
Put a number on likelihood, since words like "possible" and "likely" mean different things to different readers. Give a number, such as a 60 percent chance of winning, and say what would move it. Then run the math. If each side pays its own lawyers, a 60 percent chance at $200,000 is worth about $120,000. With $150,000 in fees to get there, the case you'd probably win is a losing bet, even before you count management's time or ask whether the other side can pay.
Most of the inputs aren't in any file you'd hand a tool: how long cases take to reach trial in your county, how your judge handles motions like yours, how your witnesses will hold up, and how the other side's lawyer tends to play it. Those come from experience and from outside counsel who practice there.
Think past the first move, too. A lawsuit means discovery, so the other side reads your executives' emails and texts, and your sales leaders sit for depositions. Cutting off one distributor sends a message to the rest. AI answers the question it's given, so finding the second and third consequences falls to you.
Finally, commit to a recommendation. New lawyers often stop at "here are the risks; it's a business decision." It usually is the business's decision, but the person making it wants to know what you would do. The comment to Rule 4-2.1 begins, "A client is entitled to straightforward advice expressing the lawyer's honest assessment." Say what you would do and why. If they choose differently, help them lower the risk.
Using AI without the blind spots
Spend the time it saves with your clients. Let it take the first draft and the first pass at research, and use the hours for meetings.
Before anything goes out, add what the tool couldn't know: the history, the goal, the people, and your recommendation. If your answer would be the same for any company that asked, it isn't finished.
Write it in your own words. Internal clients notice a pasted answer, and once they think you're passing along the tool's output, they'll go to the tool directly.
Find a senior lawyer to test your judgment against. AI won't tell you when your instinct is wrong for this company.
If you're new to an in-house role and want to talk through any of this, I'm always happy to chat.
This article is general information, not legal advice, and reading it does not create an attorney-client relationship. Sean Jordan Law PLLC, Winter Springs, Florida. Sean Jordan is admitted to practice in Florida.
