Where I'd actually start with AI in a one- or two-lawyer practice

The trap almost every solo practice falls into

I've built systems for maybe a dozen one- and two-lawyer practices at this point, and they almost all arrive at the same starting point: a graveyard of tool subscriptions. A drafting assistant they tried for a month. A research tool they pay for but rarely open. A chatbot bolted onto their intake form that nobody trusts enough to actually rely on. None of it adds up to anything, because none of it was built around how the practice actually spends its week.

The instinct to buy a tool is understandable. Tools are easy to demo and easy to cancel. But a solo practice doesn't have the luxury of running five half-adopted experiments in parallel. You have maybe four or five hours a month of attention to spend on this, total, once you account for actually practicing law. That attention needs to go toward one thing that works, not five things that might.

So before I'll even talk to a solo lawyer about which model or which vendor, I ask them to map their week. Not their aspirations for their week, the actual week: what comes in, what they do with it, and where the hours go. That map is the whole plan. Everything else follows from it.

Map the week before you touch a tool

1. 1. List the matter types that make up 80% of your docket

   For most solo practices this is two or three categories. An immigration solo I built a system for had it down to family-based petitions and employment-based petitions, full stop. Everything else was a rounding error. Know your two or three before you do anything else.

2. 2. Track where a new matter actually loses time

   Is it the intake call and the back-and-forth to get documents? Is it the first draft of a standard filing? Is it chasing a client for the fifth time for the same three documents? Write down the actual bottleneck, not the one that sounds most impressive to fix.

3. 3. Separate judgment work from assembly work

   Judgment work is where you're actually deciding strategy or reading a client's situation. Assembly work is where you're reformatting, re-explaining, or re-checking something you've done a hundred times before. AI belongs in the second bucket first, not the first.

4. 4. Pick the single highest-volume assembly task

   Not the most interesting one. The one that recurs the most often. That's where a small system pays for itself the fastest, and where you'll get the repetitions you need to trust it.

Why I almost always start with intake and triage

Intake is unglamorous, which is exactly why it's the right place to start. It's high volume, it's low stakes in terms of the finished work product (you're not filing anything based on it), and it's where a solo practice bleeds the most hours to context-switching. Every new inquiry means stopping what you're doing, reading an email or listening to a voicemail, and deciding whether it's a fit, what it needs, and what to ask for next.

The first computed module I build for a solo practice is almost always an intake triage system: something that reads an inbound inquiry, checks it against the practice's actual matter criteria, and produces a structured summary plus a recommended next step. Not a decision, a recommendation with the reasoning shown, so the lawyer is still the one who says yes or no.

The reason this works as a first project has nothing to do with AI being especially good at intake. It's that the downside of a wrong answer is small (you re-read one inquiry) and the volume is high enough that you get real signal on whether the system is helping within two or three weeks, not two or three months. That fast feedback loop is what builds the trust you need before you touch anything closer to the actual legal work.

The second project: one drafting module, not a drafting practice

Once intake is running and you trust it, the next move is one computed module for one standard document type in your highest-volume matter category. Not a general-purpose drafting assistant, one specific document: the standard demand letter, the standard petition cover, the standard engagement letter variant you use ninety percent of the time.

I build these narrow on purpose. A general drafting tool has to be good at everything, which means it's mediocre at your specific thing. A narrow computed module only has to be good at the one document you feed it constantly, and it can be built with your actual precedent language, your actual clause preferences, and your actual client-facing tone baked in from day one.

Get your evidence discipline right before you scale up

This is the part solo practices skip and regret. Any system that touches facts, whether that's case details, client history, or prior filings, needs a rule about where it's allowed to pull from and how it shows its sources. I call this evidence anchoring, and I've written a full guide on how I build it, but the short version for a solo practice is: never let a system state a fact it can't point back to in your own file.

This sounds like extra work up front, and it is. It's also the difference between a system you can defend to a bar complaint and one you can't. Build the citation habit into the first module, even the boring intake one, so it's already a habit by the time you're building something with more exposure.

Give it six weeks before you judge it

The most common mistake I see after a solo practice does build something reasonable is abandoning it in week two because it feels slower than the old way. It usually is slower in week two. You're learning to write a good prompt, learning what the system is bad at, and building the habit of checking its output instead of just trusting your own draft instinct.

Six weeks is roughly what it takes for that friction to disappear and for the time savings to become obvious rather than theoretical. I ask every solo lawyer I work with to commit to that window before we talk about whether something is working or not.