AI is being marketed to law firms as either transformative or dangerous. Depending on the headline, it will replace associates or expose firms to malpractice risk. I find both narratives unhelpful.
If I were designing an AI assistant for a small, local law firm, perhaps a 3–10 attorney practice handling personal injury or family law, I wouldn’t start with disruption. I would start with containment. The goal wouldn’t be to automate legal judgment. It would be to reduce friction around information while preserving reliability and trust.
Here’s how I would approach it.
1. Start With Internal Knowledge, Not Client-Facing Automation
Small firms accumulate years of operational intelligence: intake checklists, filing procedures, template letters, court-specific nuances, billing policies, prior motion language. The problem isn’t a lack of knowledge. It’s that knowledge is often scattered.
Before building anything public-facing, I would design an internal knowledge assistant trained exclusively on firm-approved documents. Its job would be narrow and controlled: retrieve and summarize what the firm already knows.
That assistant could help staff quickly answer questions like:
- “What’s our intake checklist for a car accident case?”
- “What’s the filing deadline for this type of motion in King County?”
- “Where is the latest version of our demand letter template?”
- “What documents do we require before filing this claim?”
The architecture would be simple and defensible:
- Secure document repository
- Retrieval-based AI (not open internet generation)
- Source citation for every answer
- Role-based access controls
- Logged queries for accountability
Notice what this system does not do: it does not invent legal advice. It does not reason about case law. It retrieves internal firm guidance and presents it clearly.
For many small firms, simply reducing the time spent searching and clarifying internal procedures would deliver measurable value.
2. Add Assistive Layers Around Intake and Drafting
Once internal retrieval proves stable, I would cautiously extend AI into areas where it can reduce cognitive load without replacing judgment.
Client intake is a good example. Law firms routinely receive long, unstructured narratives describing incidents. Someone must read them, extract key facts, and determine whether the matter fits the firm’s focus. An AI assistant could summarize inbound intake messages and structure the essentials — dates, locations, type of incident, missing information — before a human reviews the case.
Similarly, AI can assist in drafting routine communications. It can generate first drafts of client updates, summarize case timelines, or convert internal notes into structured memos. But it should operate under clear constraints.
I would explicitly limit its role to:
- Summarizing intake submissions
- Drafting routine client communications
- Preparing internal case summaries
- Reformatting and organizing existing information
And I would explicitly avoid:
- Autonomous legal advice
- Drafting court filings without full review
- Strategic case recommendations
- Direct client interaction without human oversight
The distinction matters. In this model, AI behaves like a supervised junior associate. It accelerates the mechanical portion of the work but never removes responsibility from the attorney.
Every output would require review. Every decision would remain human.
3. Design for Risk Containment, Not Novelty
The most important design principle in a legal environment is reversibility. If something goes wrong, the firm must be able to step back instantly.
That means:
- Clear internal disclaimers about AI limitations
- Mandatory human review for substantive outputs
- No unrestricted internet-connected generation
- Tight data access controls
- Periodic audits of system behavior
The temptation with AI is to push toward autonomy because it feels advanced. For a local law firm, that is rarely the right starting point. Reputation and client trust are fragile. An efficiency gain that introduces risk is not a gain.
If I were piloting this system, I would begin only with internal knowledge retrieval and measure whether it reduces time spent searching for documents and clarifying procedures. If that proves reliable, I would cautiously add intake summarization. Drafting support would come later — and only within defined guardrails.
The opportunity for AI in small law firms is not replacement. It is refinement. By reducing friction around information, attorneys can focus more energy on strategy, negotiation, and client relationships — the work that actually requires expertise.
The future of AI in legal practice may eventually expand. But if I were building an assistant for a local firm today, I would design something modest, controlled, and accountable.
Not a substitute for judgment.
A tool that makes the firm’s existing intelligence easier to access — and safer to use.