Don't Buy Legal AI!!! Until You Know This...

I have spent the last two years, helping attorneys understand AI and helping AI platforms to understand that that unlike me, attorneys don’t have the luxury of spending as much time as they want geeking out on all things AI.

I've helped attorneys understand the good, the bad and the ugly, when it comes to AI (psst, there's a lot of bad and ugly out there!). I’ve talked and partnered with dozens of AI founders, developers and CAIO’s (which is not Italian for good bye!) and watched as their business development directors spoke above their potential clients, seen the collective heads nod as if they understood, and knew that the decision would be to do nothing, because that was precisely what they understood.

But to my friends in the legal profession, this is no longer a choice. You must start using AI. But before you rush off to buy that box of AI Panacea from Radio Shack, you absolutely must understand some basics or you'll get screwed.

So here's you're AI Decoder Ring, broken into three parts: Basics & Safety, APIs, and Integration. Within a few minutes, we'll turn “What the hell is that?” into “Ah yes, of course.”

Part 1: The Basics & Safety — Protect Your Clients and Your Career

Closed vs. Open Models

What it is:

Closed models—like GPT-4 and Gemini—are black boxes: powerful but private. Open models—like LLaMA or Mistral—share their code and training data (at least partially).

Why it matters:

ABA Rule 1.6 and breach of confidentiality, that’s why. Some idiots uploaded the Samsung Master code to Chat GPT when it was an open model platform (it has since been converted to closed, btw). It was then leaked to the world. Oops. Guess who didn’t get a Christmas bonus. So you need to know what kind of model you’re marrying. Because once the honeymoon’s over, it’s your malpractice insurance on the line.

RAG (Retrieval-Augmented Generation)

What it is:

RAG pulls in your actual firm docs—contracts, memos, filings—before the AI spits out a response. In other words, it uses real-world data.

Why it matters:

No more made-up cases or fantasy statutes as in the humiliating Avianca Airlines case, that included 6 fictitious citations. RAG keeps your AI grounded in your data, not Reddit’s best guess. It’s the difference between a helpful associate and an improv actor in a suit.

Data Asset Creation

What it is:

Remember the movie “Class Action” with Gene Hackman (don’t worry if you don’t, it’s from 1991). Great legal thriller. In it, the AmLaw firm is required to provide the memos from an engineer about some cars that have the annoying habit of blowing up (Can you say Ford Pinto? Instead of providing the memos from that engineer, they “paper them to death” and provide a million pieces of paper – for five attorneys to go through, manually.

Why it matters:

Imagine the difference it would make if with a single keystroke, they could go through all that paper and find exactly what they were looking for. Thus it is with data asset creation. Every memo, contract and even clause of a contract that your firm has ever created or filed, can be retrieved through a simple search. Important stuff.

Guardrails

What it is:

Hardcoded bumpers that stop your AI from offering stock tips, off-topic nonsense, or legal advice that would get a first-year fired.

Why it matters:

Because “oops” isn’t a defense strategy. Guardrails reduce risk, limit liability, and keep the AI from wandering off like a caffeinated summer intern on a Slack thread. Plus “tell me about the guardrails” just sounds cool to say, when talking with vendors!

Audit Trail

What it is:

This is just what it sounds like. A clear, clickable e-trail of who prompted what, when, and with which result. To the point it will stand up to an audit.

Why it matters:

When someone asks “how the hell did THIS recommendation happen?” the answer shouldn’t be “¯\_(ツ)_/¯.” Audit trails give you receipts - and regulatory agencies and their minions love receipts.

Part 2: APIs — The Digital Plumbing You Can’t Afford to Ignore

API / API-First Strategy

What it is:

APIs let software systems talk to each other. API-first means the product was built to integrate—not duct-taped on later.

Why it matters:

If your shiny new AI tool will segregate emails into Trash / Spam, Categorize and Priority, but it won’t work with Outlook, you’ve just bought a very expensive icon for your desktop. If in order to get it to work with Outlook, you need a tech, you’ve just doubled the cost of your AI. But if it was designed to work with Outlook from the beginning, you’ve got a great tool (btw, if any DOES have an AI that will do this with Outlook, let me know!)

Event-Driven API

What it is:

If something happens, this API makes sure your AI does what it needs to do. For example, if the AmLaw dumps a million pages of documents into your inbox, the AI won’t wait for you to tell you what to do. It will immediately begin classifying, sorting and seeking out the document you actually need. By the time you’re having coffee the next morning, the one memo that will get your client a million dollar verdict, is waiting for you! This can apply to everything from a contract being signed, to being alerted that the SEC has filed a suit against your favorite client.

Why it matters:

No more manually instructing your software what to do, after something has happened. Just fast, fluid workflows that update faster than an overachieving associate on their fifth Shaken Espesso from Starbucks.

RESTful API

What it is:

This does a couple of easy and important things. First, it makes integration simpler, smoother, and way less likely to implode. Second, it allows you to just give you AI an endpoint website e.g. litigation docket and tell it what you want to know. For example, I would strongly recommend this for Talent-side Sports & Entertainment attorneys who want to know whether their favorite but troublesome client has been arrested again, at 3am in the morning.  

Why it matters:

Your IT team will thank you. Your wallet will thank you. Your heavy-drinking NFL all star will thank you.

Part 3: Interoperability — Integration Without Interruption

Modularity

What it is:

Ever see “modular” furniture? They’re like soft comfy Legos. Move into a bigger house and you can just buy more pieces. Find one piece that just isn’t comfortable or doesn’t go with the décor anymore, you don’t have to keep it.

Why it matters:

Don’t buy a monolith. Buy a toolbox. Modularity means you can scale, adapt and swap out underperformers like that partner who promised a $5M book and hasn’t produced $100K in his first six months.

Frictionless GTM Execution

What it is:

GTM = Go-To-Market. “Frictionless” means fast onboarding, super user friendly so that you have a higher “adoption” rate (meaning people will actually USE what you’ve paid for!), and maybe even a support person who answers your calls or email without just sending you a 67-page .pdf that you’ll never read.

Why it matters:

If your firm needs 6 weeks of training just to find the login screen, adoption will die a slow, expensive death. Fast setup, easy-to-use = faster savings or ROI. That’s the goal of every investment for your firm.

Scalability

What it is:

Can the tool handle more matters, more users, more complexity—without falling apart like the attorney who turned himself into a cat avatar in a trial during COVID?

Why it matters:

What works for your 10-lawyer boutique might buckle under your 100-attorney expansion. Buy tech that grows with you, not against you.

Final Verdict:

As mentioned in my last article, AI isn’t optional. It’s quickly approaching the ABA Rule 5.1 in that, if you're not using it, you're charging unreasonable fees. BUT

If you don’t understand the vocabulary, you’ll end up buying flash over function. Keep this list handy or better yet, learn what the heck you just read! That way, the next time a vendor starts tossing around jargon, you won’t nod like the Million Mindless Management Committees who have been left scratching their heads and either passing on gold or purchasing garbage.

Better yet, you can be the one using the jargon and enjoying the blanks stares of all your newly impressed tech-less Boomer friends and partners! Not that attorneys like to be the smartest person in the room...

Frederick Shelton is a Market Advisor and Consultant to law firms, legal MSO's and funds on subjects which include legal AI, ABS models, MSO's and M&A. He can be reached at fs@sheltonsteele.com