A firm owner sent me a Lawmatics build to “just check before we go live.” It was good. Suspiciously good. Every stage labeled, every automation mapped, a lead-scoring rubric with criteria I’d never seen her use.
I asked where the criteria came from. She said, a little sheepishly: “I described my firm to ChatGPT and asked it to design the intake.”
Here’s what was strange. The design was clean, professional, and completely generic. It scored leads on things her firm didn’t actually care about, and ignored the one factor that decided almost every case she took. The AI hadn’t designed her system. It had designed a system — a plausible one — and stapled her name to it.
That’s the trap I want to talk about. Because after the last few months of AI announcements, the question I get has shifted. It used to be “should I turn on the AI?” Now it’s “can AI just design the whole thing for me?” The honest answer is yes — use it — but not the way most people are about to.
The question underneath the question
When I wrote about when to turn on Lawmatics’ AI, the point was simple: AI multiplies your process, it doesn’t repair it. Design first, tools second.
The pushback is fair: “But I’m using AI to design. Doesn’t that flip the order?”
It doesn’t. It just moves the same question up a level. When AI runs your intake, it multiplies whatever system is underneath. When AI helps you design the intake, it multiplies whatever judgment is underneath. Feed it clear thinking and it makes you faster. Feed it a blank space where your judgment should be, and it fills that space with something confident and average.
The tool is the same. What changes is what you’re handing it to multiply.
What AI is genuinely good at when you design
I use AI when I build systems. I’m not precious about it — it’s very good at a specific set of things, and pretending otherwise is just slower.
It turns a mess into a first draft fast. Take a rambling description of how intake “sort of works right now” and lay it out as stages you can actually look at. It surfaces the steps you forgot and the edge cases you’ve been avoiding — the “what happens when a lead comes in at 6pm on a Friday” nobody wants to think about. It writes the first version of an email, a form, a checklist, so you’re editing instead of staring at a blank page.
Notice what it’s doing in all of those: accelerating work you already know how to judge. You look at the draft and you can instantly tell if it’s right, because the standard lives in you. That’s AI as a co-designer. Used that way, it’s a real multiplier.
What it can’t do — and where firms get hurt
Here’s the line. AI can draft your system. It can’t decide what your system is for.
It doesn’t know that you take the hard custody cases other firms turn down, and that’s your whole reputation. It doesn’t know that one referral source is worth chasing at any hour and another never converts. It doesn’t know the quiet rule you’ve never written down but apply on every call. Ask it to invent those, and it will — smoothly, plausibly, wrongly.
I watched a firm run an AI-designed scoring setup for six weeks before anyone noticed it was quietly tagging their best referral source — a small network that sent them four signed cases a month — as low priority. The model had no way to know those leads were gold. Nobody had told it. It optimized for what looked reasonable, and reasonable was wrong.
That’s the same failure I see in almost every audit, just wearing a new outfit. The old version was buying a CRM and hoping the tool would supply the process. The new version is opening a chat window and hoping the model will supply the judgment. Both outsource the one thing that can’t be outsourced: what a good client looks like to you.
And this version hides better, because the output looks like expertise. A blank Lawmatics account obviously needs work. An AI-generated intake design looks finished. You’re far more likely to ship it without noticing it was never really yours.
How I actually use AI when I design a system
Here’s what that looks like in a real consultation — because I use AI constantly, just never to decide anything.
When a firm already has something built, I put it in front of AI to read faster than I could alone. A Lawmatics account that’s been added to for two years is a sprawl: dozens of automations, half of them orphaned, stages nobody remembers naming. AI helps me map it in minutes — what connects to what, where a lead can get stuck, which automation fires twice and which never fires at all. It surfaces the blockers and the gaps. I decide what they mean.
When a firm has nothing built yet, I use it the other direction. We talk through how the firm actually runs, and AI helps me lay it out as a picture — where the weak points are, where everything depends on one person remembering, and, just as usefully, where they’re already strong and shouldn’t be touched. Firms rarely see their own strengths. Naming them is half of a good design.
Either way, AI does the seeing-fast; I do the judging. It shows me the bottleneck, the gap, the soft spot — then I decide what to protect, what to rebuild, and what to design from scratch. Only then does it help me build the new version quickly, on top of decisions we’ve already made.
The AI never sets the criteria. It executes against criteria we’ve already decided — the same principle as treating intake as a decision engine, not a form. The decisions are yours; the tool just runs them.
In that order, AI has made the work genuinely faster. In the other order, it just gets you to the wrong system sooner.
The one thing to check this week
If you’ve used AI to design any part of your firm — an intake flow, a scoring rule, a follow-up sequence — open it and ask one question: could I have written the judgment behind this myself, or did the AI fill that in for me?
If you can point to where your thinking lives in it, good. Keep going — you’re using it right. If you can’t, that’s not a reason to throw it out. It’s a reason to step back and put your judgment where the AI guessed.
Because “AI-native” was never going to mean the AI does the thinking. It means your thinking is clear enough that the AI can move fast without inventing the parts that matter.
Getting a firm’s real judgment onto the page — then using AI to build fast on top of it — is the order we work in on our Customized AI Solutions engagements. If you’ve got a design you’re not sure is really yours, that’s the gap worth closing before it goes live.
So before you ship the system AI helped you build: is it running your judgment — or just something that sounds like it?
