---
title: "The AI-native law firm: from 15 to 100 cases per case manager without losing empathy"
description: "An AI-native immigration law firm went from 15–20 to 80–100 active cases per case manager: AI compiles the case files and people take care of clients."
url: https://addabra.com/en/blog/ai-native-law-firm
lang: en
date: 2026-10-15
category: "Industries"
keyword: "AI-native law firm"
publisher: Addabra
alternate: https://addabra.com/blog/despacho-legal-ai-native
---

# The AI-native law firm: from 15 to 100 cases per case manager without losing empathy

> An AI-native immigration law firm went from 15–20 to 80–100 active cases per case manager: AI compiles the case files and people take care of clients.

Monday, 8:15 a.m. A case manager at an immigration law firm opens her inbox: 43 new emails, three clients asking "How is my case going?", a piece of evidence that was scanned sideways and a Thursday deadline for a petition of more than 400 pages. She has 18 active cases and feels she can't take on a single one more. While she builds indexes and numbers exhibits, a client is crying on the phone because she's afraid for her family. The case manager listens for five minutes, feeling guilty, eyes on the clock.

This is the reality at many legal practices. And it's the starting point for one of the systems Addabra built: an **AI-native law firm** in immigration where each case manager went from handling 15–20 active cases to 80–100, without clients feeling any less cared for. Quite the opposite.

## The bottleneck isn't the law, it's the case file

When we talked to the firm, the diagnosis was clear. Case managers weren't overwhelmed by the legal complexity of their cases, but by everything surrounding the case file:

- **Compilation.** An immigration petition can run past 400 pages: forms, support letters, translations, documentary evidence, indexes and exhibits that must be cited correctly.
- **Timelines.** Reconstructing the client's history with exact dates from scattered documents.
- **Deadlines and SLAs.** Every case has government deadlines and internal response commitments. Missing one has serious consequences.
- **Follow-up.** Requesting missing documents, chasing signatures, updating the client on every step.

All of that work is essential, but it isn't judgment work. And it was eating up most of the day.

## What it means to be an AI-native law firm

An AI-native firm isn't one that "uses ChatGPT." It's one that redesigned its operations around a single question: what does a person need to do, and what can a well-designed system do?

The firm's answer was blunt, and we adopted it as a design principle:

> You can automate compiling a case file. You can't automate caring about the client.

With that, the split looked like this:

| The system does | People do |
|---|---|
| Compile petitions of hundreds of pages with indexes and exhibits | Set the strategy for each case |
| Build timelines from the client's documents | Review and sign every document before it's filed |
| Track deadlines and SLAs, and alert well in advance | Reassure the client and explain what comes next |
| Detect missing documents and request them | Support clients through hard moments with empathy |
| Draft letters and status updates | Decide when a case needs a different approach |

## How the system is built

Behind that split is an agentic system with several pieces working together:

1. **An intake agent** that receives the client's documents, classifies them, names them by a standard convention and flags what's missing.
2. **A timeline agent** that extracts dates and facts from each document and builds a verifiable timeline, with a reference to the source page.
3. **A compiler agent** that assembles the full petition following the firm's template: order, indexes, exhibit numbering and cross-references.
4. **An evaluator**, separate from the compiler, that checks consistency: that dates match across documents, that every cited exhibit exists and that no required piece is missing.
5. **A deadline agent** that monitors each case's SLAs and alerts the case manager days in advance, not on the due date.
6. **A communications agent** that prepares client updates in the client's own language, which the case manager reviews before sending.

Everything rests on a per-case memory: the client's history, their documents, their previous questions and the decisions made. When the case manager opens a case, she sees the full context in a minute.

## Human sign-off is non-negotiable

At a law firm, a mistake isn't an inconvenience: it can cost a family its future. That's why every outgoing document passes through a gate: an approval checkpoint that only the person with the right role can open. The case manager reviews the compilation, the attorney signs the petition, and the system logs who approved what and when. We go deeper on this design in [human approval by role](https://addabra.com/en/blog/human-approval-by-role).

That structure doesn't slow the team down. Just the opposite: because the system already did the compilation and the evaluator already checked consistency, human review focuses on what matters: strategy and arguments.

## From 15–20 to 80–100 active cases per case manager

The most visible result is capacity. Each case manager went from handling 15–20 active cases to 80–100. But the number doesn't tell the whole story.

What changed in a case manager's day:

- **Fewer hours assembling, more hours thinking.** The first draft of the petition is ready when she opens it.
- **Fewer scares.** Deadlines raise alerts in time and nothing lapses because someone forgot.
- **More time with clients.** The call from the worried client is no longer taken with one eye on the clock.

For the firm, that means growing its caseload without growing headcount at the same pace, a model we explore in [grow without hiring](https://addabra.com/en/blog/grow-without-hiring).

## Illustrative example: one case, end to end

Picture a new client arriving with a box of documents. Before, the case manager spent days scanning, sorting and building the timeline. Now the client uploads her documents and within a few hours the system returns a sourced timeline and a list of what's missing. The case manager meets with her, reviews the timeline, clarifies two dates and works out the strategy with the attorney. The compiler assembles the petition; the evaluator flags an inconsistency in an entry date; the case manager fixes it, the attorney signs and the petition is filed ahead of the deadline.

Throughout the process, the client spoke with a person every time she needed to.

## Common mistakes when bringing AI into a law firm

- **Starting with what's most visible instead of what's heaviest.** A website chatbot impresses; compiling case files changes the operation.
- **Letting the same model generate and review.** Verification has to be independent.
- **Ignoring permissions.** Each agent should see only what it needs from the case file.
- **Measuring only speed.** Also measure errors caught before filing and client satisfaction.

## What to do on Monday

Ask your case managers to log for one week where their time goes: compiling, finding documents, chasing deadlines, talking to clients. The ratio between case-file work and judgment work will tell you how much you stand to gain.

At Addabra we build custom systems for law firms and in-house legal teams: we diagnose the process, design the agents, the gates and the per-case memory, and support them in production. If the contract side interests you, see also [AI contract review](https://addabra.com/en/blog/ai-contract-review).

## Frequently asked questions

### Does AI make legal decisions at the firm?

No. AI compiles, organizes, drafts and tracks deadlines. Case strategy, final review and signature belong to attorneys and case managers, and no document goes out without human approval.

### How is sensitive client information protected?

Each agent can access only the case files and data it needs for its task, with least-privilege permissions and a log of every access. Client information is never used for anything outside their own case.

### Does this only apply to immigration firms?

The case we describe is immigration, but the pattern works for any practice with large case files, strict deadlines and clients who need support along the way: employment, family, insurance or intellectual property.

## About Addabra

Addabra is an AI-software-first software factory that designs and builds custom agentic systems for companies: one agent per process, a shared business memory, orchestration, human approval by role, auditing and cost control. Free diagnosis: https://addabra.com/en/blog/ai-native-law-firm#contacto