Start AI in this order: use a general assistant like Claude daily for 30 days, build a prompt library for your practice area, connect your data with MCPs, automate one workflow end to end, then build your own tools. Legal-specific software comes last, after the habits exist.
I own 10 law firms. My flagship handles lemon law cases in all 50 states with 62 staff. I work about 30 hours a week and touch zero daily operations.
That is not a personality trait. It’s the result of adopting AI in a specific order, and watching hundreds of other lawyers adopt it in the wrong order.
The wrong order looks like this: a lawyer hears about AI, buys a $400/month legal AI subscription, uses it four times, concludes AI is overhyped, and goes back to doing everything manually. The subscription auto-renews for a year. I’ve watched this exact sequence play out dozens of times in my lawyer communities.
The right order costs $20 in month one. Here it is.
| Stage | What you do | Cost | Time to value |
|---|---|---|---|
| 1 | Use Claude or ChatGPT for every drafting task, 30 days | $20/month | Week one |
| 2 | Build a prompt library of 15 to 25 tested prompts | $0 | 2 to 3 weeks |
| 3 | Connect AI to your data with MCPs | $0 | Under 15 minutes per connector |
| 4 | Automate one workflow end to end | Varies | Month two |
| 5 | Build custom tools with Claude Code | $100 to $200/month | When no tool fits |
Stage 1: Use a general AI assistant every day for 30 days
Before you buy anything legal-specific, get a paid Claude or ChatGPT account and force yourself to use it for every drafting task for 30 days. Client emails. Letters. Summaries of opposing counsel’s nonsense. Explaining a fee agreement in plain English.
The point is not that the AI writes better than you. Early on, it won’t. The point is building the reflex of asking AI first. Lawyers who skip this stage never develop an instinct for what AI is good at, so every later tool disappoints them.
My rule for new staff: if a task involves producing words and takes more than 10 minutes, you draft it with AI first. No exceptions for the first month. After that, they never need the rule because they’ve seen the math.
Start with Claude. The writing is stronger for legal work and it handles 200-page documents without choking.
Stage 2: Build a prompt library for your practice area
A blank chat box is where AI enthusiasm goes to die. The fix is a prompt library: 15 to 25 tested prompts for the tasks your firm repeats.
At my lemon law firm, intake staff don’t improvise. They have a prompt that takes raw client notes and returns a structured case summary with vehicle details, repair history, and a strength assessment. Same input format every time, same output format every time. That consistency is what makes AI usable by a team instead of one enthusiastic partner.
Write prompts for: your top three client email situations, your most common document summary, your intake notes format, and one drafting task you personally hate. That’s a real library. Copy mine from the prompt guide if you want a head start.
Stage 3: Connect AI to your actual data
This is the stage most lawyers never reach, and it’s where the leverage lives. A chatbot that can’t see your calendar, your email, or your case management system is a smart intern locked in a closet.
MCPs (Model Context Protocol connectors) let Claude read and act on your real systems: Gmail, Google Drive, Slack, your CRM, and yes, Clio. Once connected, “summarize every email from opposing counsel on the Martinez matter this month” is a 10-second task instead of an hour of scrolling.
Setup is under 15 minutes per connector and requires no code. The MCP guide lists the ones worth connecting first.
Stage 4: Automate one workflow end to end
Pick one workflow that happens at least weekly and automate the whole thing. Not 80% of it. All of it.
At CPLG, document collection used to be a paralegal chasing clients by phone for repair orders and purchase contracts. Now an automated sequence sends the request, reminds at day 3 and day 7, reads what the client uploads, checks it against what the case needs, and only involves a human when something’s actually wrong. The paralegal handles exceptions, not routine.
Your first automation should be smaller: intake follow-up, review requests, or a weekly status email to clients. One complete workflow teaches you more than ten half-automated ones.
Stage 5: Build what doesn’t exist
Here’s what changed in the last year: when no tool fits, you can build your own by describing it in English. Claude Code turned my firms from software buyers into software owners. A dashboard tracking sign-ups across all 10 firms, a document review system for lemon law files, an intake scoring tool: all built by AI, all owned by me, no subscriptions.
A restaurant owner in one of my communities needed dynamic pricing software, was quoted $10,000, and built it himself with Claude Code in an afternoon. The same math applies to every legal-specific tool you’re renting. That’s stage five, and it’s more reachable than you think.
Where the legal AI tools fit
Notice that “buy legal AI software” isn’t a stage. Tools like CoCounsel, Spellbook, and EvenUp are real and some are excellent. But they pay off only after your firm has the habits from stages 1 and 2. A firm that prompts well gets 5x more out of any legal AI tool than a firm that bought the tool first.
When you’re ready, the catalog on this site covers the legal-specific tools one by one, with what they cost and who they’re actually for. AI legal research tools and AI demand letter tools are the two categories with the clearest payoff for most firms.
The 60-day version
This site publishes one resource a day for 60 days, in roughly the order above. Subscribe below and follow along, or skip ahead through the full catalog. Either way: open Claude today and draft one real client email with it. That’s day one, and it costs nothing.