The Attorney’s AI Workflow: Prompt Patterns, Live Demos, and the Verification Habits That Keep You Off the Sanctions Docket

The on-ramp hour, built for attorneys who have opened a chat window and been underwhelmed. The session establishes what large language models actually do and where they fail, in plain terms, then builds the anatomy of an effective legal prompt one element at a time: role, context, task, format, constraints. Three live demonstrations put that structure to work on the tasks that return time immediately, each ending with its prompt pattern displayed full-screen and reproduced in the prompt library. No AI background assumed at any point.

For lawyers who already use AI every week and want to see what it can do with a full matter file. The session runs about 15 demonstrations drawn from family law, construction and business litigation, estate and trust, and insurance coverage. Each one starts with a realistic document set and ends with work product a supervising attorney would actually review. You’ll watch AI build a sourced chronology from a deposition and its exhibits, flag where the testimony and the documents disagree, and carry that into a draft declaration. You’ll see a signed stipulation turned into a docketing list with every deadline tied to the language that triggers it. And you’ll see the supervision habits that make this work safe on a client matter, from review stops between steps to a verification pass before anything leaves the room. Every file is fictional.

The session moves through chained workflows, in which a deposition and its exhibits become a sourced chronology, then an issue outline, then a draft declaration or motion section, with a review stop at each handoff. It demonstrates working across a document set, synthesis across a full matter file, including a contradiction register that shows exactly where the witnesses and the record part ways and exception-first review, with risk-spotting on an incoming contract and a structured critique of an opposing draft, problems surfaced first and every finding tied to its page. On orders and deadlines, a stipulation, MSA, or judgment is converted into a calendaring list with the trigger language quoted and each date held as unverified until a person checks it.

Attendees also keep the tools and patterns behind the demonstrations: interactive settlement models and an intake summarizer built in conversation with no code, plus Projects and Skills that hold format and constraints steady from one matter to the next. The session closes with catching failure mid-workflow, spotting the signs that output is slipping, like a deadline that lands on a Sunday or a citation nobody can find and recovering without starting over.

A working ethics hour organized around the path from AI output to filed documents. It walks the pre-filing verification protocol step by step, citations, quotations, factual assertions, then moves to the confidentiality settings that determine what a model remains under Rule 1.6, and the supervision policies Formal Opinion 512 and Rules 1.1, 5.1, and 5.3 expect once associates and staff touch these tools. The current sanctions landscape supplies the stakes, not the syllabus. (1 Ethics Credit)