The five AI agents every law firm should build, in order: an intake specialist, a paralegal, a marketer, an operations manager, and a strategist. Each one starts as a Claude Project loaded with your documents, no code required.
Most lawyers use AI like a search engine with better manners. Open a chat window, type a question, copy the answer, close the tab. Useful, but that’s a chatbot.
An agent is different. An agent acts. It watches for a trigger, does multi-step work on its own, and uses your actual systems: your lead forms, your templates, your case data. A chatbot answers when you ask. An agent qualifies the lead that came in at 2 AM, drafts the demand letter from your template, and flags the case that’s been sitting in document collection for 30 days, without you opening anything.
I own 10 law firms. My flagship handles lemon law in all 50 states with 62 staff, and I work about 30 hours a week with zero daily operations. That structure runs on agents. Here are the five to build, in the order that pays fastest.
What makes something an agent
Three ingredients. It responds to triggers instead of waiting for you to type. It chains steps together: read, decide, draft, route. And it touches your real systems instead of living in a blank chat box.
You don’t need all three on day one. Every agent below starts as a conversation you can have this week, then graduates to full autonomy. The order matters, so start at the top.
| Build order | Agent | Job | Difficulty |
|---|---|---|---|
| 1 | Intake specialist | Scores every lead in seconds | Easy |
| 2 | Paralegal | First drafts from your templates | Easy |
| 3 | Marketer | Ad variants plus weekly performance reads | Medium |
| 4 | Operations manager | SOPs and time-leak audits | Medium |
| 5 | Strategist | Pressure-tests growth ideas against your numbers | Hard |
Agent 1: the intake specialist
Upload your entire intake process: qualification questions, case criteria, red flags, and your actual accept and reject reasons from past leads. Then give it this instruction: “Build me a scoring system. When a lead comes in, qualify it against my criteria, score the case, tell me sign, decline, or refer, and explain your reasoning.”
That prompt alone produces something most firms have never had: consistent intake decisions with written reasoning. Two different intake staff will score the same lead differently on a Friday afternoon. The agent won’t.
Then connect it to your lead forms so it runs the moment a lead hits, 24/7. Speed-to-lead goes from hours to seconds, and speed-to-lead is the single biggest controllable variable in signing rate. The lead you call back in four hours already talked to two other firms. The full intake guide covers the wiring.
Agent 2: the paralegal
Feed it your templates: demand letters, retainers, client emails, the motions you file over and over. Add your firm’s voice and real citations from your own past work, not whatever the model invents.
First drafts now take minutes. An attorney reviews in 10 minutes instead of drafting for 4 hours. That review step is not optional, and it never goes away. What goes away is the blank page.
The mistake here is feeding it generic samples from the internet. The agent is only as good as the documents you give it. Your best demand letter from last year teaches it more than any prompt engineering trick. For demand letters specifically, I compared the dedicated tools here, and the prompt guide covers the drafting patterns that make this agent sharp.
Agent 3: the marketer
This one generates ad variations, hooks, creative concepts, and audience recommendations. More importantly, it analyzes which ones performed and iterates. Give it your ad account data every week and ask what to kill, what to scale, and what to test next.
Marketing agencies charge law firms $5,000 a month largely for this loop: produce variants, read results, adjust. An agent runs the same loop daily for the cost of a subscription. You still make the calls on spend. The agent makes sure you’re choosing between good options instead of staring at a dashboard.
Agent 4: the operations manager
This agent builds SOPs, maps workflows, and finds where the firm leaks time and money. Feed it your process documents, your org chart, and honest descriptions of how work actually flows, then ask: “Show me the highest-leverage tasks I should focus on and the low-leverage work to automate or eliminate.”
Every firm I’ve audited has a version of the same finding: senior people doing $20-an-hour work because nobody ever wrote down who should do what. The operations agent writes it down, keeps it current, and tells you what to hand off next. Mine flagged that our document-collection follow-ups were manual across 62 staff. Automating that one sequence bought back hundreds of hours a month.
Agent 5: the strategist
The one most lawyers skip. Also the most important.
Load this agent with your numbers: sign-ups, conversion rates, cost per lead, settlement timelines, revenue by case type. Then hold it to a standard most advisors never meet: “What specific actions change which numbers, by how much, in what timeframe? Run every idea through a what-would-it-actually-take analysis, and don’t tell me what I want to hear.”
Every growth idea sounds good in a partner meeting. The strategist agent forces each one through the same filter: which metric moves, how far, how fast, at what cost. Half my “great ideas” die in that analysis, and they deserve to. The ones that survive get resourced with actual conviction.
The warning that comes with all five
These agents amplify execution speed. They do not replace thinking quality.
I’ve watched lawyers automate a broken intake process and wonder why signings didn’t move. The agent did exactly what it was told, faster, at scale. The process was the problem, and now the broken process runs 24/7.
The tools aren’t the bottleneck. Your understanding is. AI without direction is worthless. AI with elite direction builds a firm that runs while you sleep. The five agents above are worth building precisely because each one forces you to write down what you actually know: your intake criteria, your best documents, your real numbers. Firms that can’t produce those documents don’t have an AI problem.
How to actually build these
Simpler than it sounds. Each agent starts as a Claude Project: a set of instructions plus your documents, no code. That version works today and covers most of the value.
Graduate to Claude Code plus MCPs when you want triggers and system access: the intake agent reading your lead forms directly, the paralegal pulling case facts from your CRM, the strategist querying live numbers instead of last month’s export. That’s the difference between an assistant you consult and an employee who shows up on their own.
Agents are stage four of the roadmap, so if the Claude Project version already feels like a stretch, back up a stage and build the foundation first.
Bottom line
One agent, built this week, beats five agents planned for next quarter. Start with intake, prove the scores match your judgment, then move down the list.
Build agent 1 this week
Not all five. One. The intake specialist is first for a reason: you already have the intake criteria document, or your intake team’s collective memory of one, and that document is 80% of the work. Write it down, load it into a Claude Project with the scoring instruction above, and run your next 10 real leads through it before you touch anything else. You’ll know within a week whether the scores match your judgment. Then, and only then, wire it to your forms.