How Can Lawyers Automate Legal Work With AI?

TL;DR

Lawyers can reduce repetitive legal work by using AI skills and specialized agents to extract case details, update databases, organize tasks, research issues, and draft filings. Paula Rojas reduced a lawsuit-intake process from about an hour to roughly five minutes, but deadline calculations and generated legal content still require human review because errors and hallucinations can reach court filings.

Transcript

five cases that come to mind of lawyers that are making filings to the courts that have hallucinations. I started with GBT just chatting away and now I'm learning CL code and vibe coding. >> I don't even believe it myself but [music] actually I built a little team of agents. >> Oh, that's so cool. >> It's about a 45 minute to 1 hour and this [music... Read More

Key Insights

• AI automation is most useful when applied first to repetitive, standardized work that consumes substantial staff time. Rojas mapped the recurring processes in her firm and selected labor-law lawsuit intake because the cases generally follow similar patterns and previously required about an hour of manual processing.
• Rojas's lawsuit-intake skill reads the complaint and initial judicial decision, identifies the filer, factual allegations, requested relief, claimed amount, financial risk, filing number, court, and hearing information, then transfers those details into the law firm's Notion databases.
• Legal deadline calculation is a difficult part of the workflow because the response is due five working days before the hearing and Chilean holidays do not count. The automation checks a holiday website, but backward counting has required repeated adjustment and verification.
• The automated intake workflow reduces an approximately one-hour process to about five minutes. That time saving matters at a litigation practice managing about 200 active labor cases, where repeated manual reading, data entry, and calendar updates consume significant capacity.
• Rojas's agent team divides work by function instead of relying on one general assistant. Separate agents handle email triage, court notifications, project breakdown, family finances, legal drafting, research, memory maintenance, and educational content for her classes about AI use by lawyers.
• The chief-of-staff agent filters information so Rojas can focus on matters that require her judgment. It monitors firm finances, prepares morning and afternoon briefings, files newsletters, reports when teammates have answered clients, and coordinates a group of specialized helper agents.
• Human accountability remains essential because AI can hallucinate content that lawyers may mistakenly include in court filings. Rojas reports that her deadline workflow still makes roughly one error in every ten lawsuits, so efficiency does not remove the need for professional review.
• Confidentiality influences the choice of AI tools for legal work. Rojas says her firm uses Claude because it provides the assurances they seek for client confidentiality, while demonstrations use a public Chilean case rather than exposing information from the firm's own clients.

Questions & Answers

Q: How can lawyers automate the intake of a new lawsuit?

Lawyers can create an AI skill that reads the complaint and the court's first decision, extracts the parties, allegations, requested relief, claimed amount, financial exposure, filing number, court, and hearing date, and then writes those details into a case-management database. Rojas's workflow also records calendar information and calculates when the response is due, reducing work previously completed through manual reading and data entry.

Q: How much time can AI save during lawsuit intake?

Rojas reports that processing a new lawsuit previously took about one hour from the client's initial notice until the case information, databases, calendars, and response deadline were fully updated. Her automated workflow completes that process in roughly five minutes. The saving is especially significant because her labor-law practice is managing about 200 active cases, making repeated manual intake a major use of staff time.

Q: What legal tasks do Paula Rojas's AI agents perform?

Her agents divide responsibilities across email triage, firm cash-flow monitoring, morning and afternoon briefings, court-notification review, productivity planning, family finances, legal drafting, legal research, memory maintenance, and educational content. One agent can identify a required filing, another can draft it, and a research agent can retrieve supporting information through the connected legal MCP. This structure gives each agent a defined operational role.

Q: How does AI help lawyers manage court notifications?

Rojas uses a legal agent that receives court-system notifications delivered to her email, visits the court website, identifies the relevant case, and determines whether the notice concerns a filing, request, or judicial decision. The agent then flags information that appears important. This reduces the burden of manually reviewing every notification while still directing consequential matters to the lawyer's attention.

Q: Why are legal deadline calculations difficult to automate?

The demonstrated deadline must be calculated backward from a scheduled hearing, with the response due five working days before that hearing. Chilean holidays do not count, so the system must consult a holiday website and extend the interval when a holiday falls within it. Rojas says this backward calculation has been the hardest part to perfect and still produces about one mistake in every ten lawsuits.

Q: Where should lawyers start when adopting AI automation?

Lawyers should begin by mapping the repetitive processes performed every day or almost every day, then select work that is low risk but capable of producing a meaningful time saving. Rojas chose labor-law lawsuit intake because the documents and required processing are usually similar across cases. She also recommends learning through trial and error instead of waiting until every skill or workflow feels fully mastered.

Q: How does an AI chief of staff support a busy lawyer?

Rojas's chief-of-staff agent triages email, monitors the firm's flow of money, and delivers two briefings each day, one in the morning and one in the afternoon. It directs her attention toward matters that require action, files newsletters, and notes when a teammate has already responded to a client. This filtering helps her distinguish important responsibilities from routine informational noise.

Q: What risks should lawyers consider when using AI for court filings?

AI-generated legal work can contain hallucinations, and the transcript references multiple cases involving lawyers who submitted filings with invented material. Rojas's own workflow also shows that automated deadline calculations can remain imperfect even after repeated tuning. Lawyers therefore need to review generated drafts, extracted case information, cited research, and calculated deadlines before relying on them or submitting anything to a court.

Summary & Key Takeaways

• Paula Rojas began learning AI after returning from maternity leave to a small family law firm with fewer available team members and a demanding schedule. Seeking more time with her children while remaining effective as a lawyer and company chairman, she taught herself through YouTube, experimentation, ChatGPT, Claude Code, and vibe coding.

• Rojas built a team of specialized agents for email triage, financial monitoring, court notifications, productivity support, accounting, drafting, legal research, and content creation. Her chief-of-staff agent produces morning and afternoon briefings, filters routine information, highlights matters requiring attention, and coordinates helpers responsible for narrower tasks.

• Her lawsuit-intake skill reads a complaint and initial court decision, extracts parties, allegations, requested amounts, financial exposure, court information, and hearing details, then records them in Notion. It also calculates the response date using Chilean holidays, although backward deadline calculations remain difficult and still produce occasional errors requiring human verification.