Questions. Comments. Ideas.
The talk works best as a conversation. Interrupt me, push back, and tell me what you’re seeing. If something I said wouldn’t hold up in your courtroom, I want to hear it.
Object loudly.
Three things I’d love to hear from the room, in the session or afterward.
How does your office actually handle discovery, charging, and trial prep?
The use cases in Part 3 were built on one office’s workflow. Yours is probably different.
Ideas, thoughts, and suggestions for ways to use AI.
The best prompts I’ve seen came from lawyers describing a task they hate, not from engineers.
What you’ve tried, good or bad.
Especially the bad. A tool that fooled a careful lawyer is more useful to the room than one that worked.
The whole talk, one line per part.
- 1The question isn’t whether the defense can use AI. It’s whether you’ll recognize it when they do.
Read their authority the way you’d read a pro se motion: pull every cite. Part 1
- 2The model is the engine. Everything useful comes through a door, and every door in is also a door out.
Prompts, search, connectors, skills, projects, agents. Part 2
- 3A skill turns one lawyer’s best process into something the whole office can reuse.
And a good one never states elements from memory. Part 3
- 4Copilot can’t install a skill, but it can follow one. Keep the rules, the step order, and the output table.
And use it to check your own citations before someone else does. Part 4
- 5Verify it with your own eyes. Enterprise accounts only. Keep the reps.
Two EDNC cases, five habits, one rule. Part 5
- 6We figure this out together.
There are no experts. Tell me what you’re seeing.
Where to go next.
The two controlling opinions
- N.C. State Bar, 2024 Formal Ethics Opinion 1, “Use of Artificial Intelligence in a Law Practice” (Nov. 1, 2024). Six inquiries with answers: competence, vendor diligence, in-house tools, signing pleadings, client disclosure, billing.
- ABA Formal Opinion 512 (July 29, 2024). Maps the existing Model Rules onto generative AI; creates no new obligations.
- Neither body wrote an AI-specific rule. Rules 1.1, 1.6(c), 3.1, 3.3, 5.3 and N.C. R. Civ. P. 11 stayed where they were.
Read the orders yourself
- Fivehouse v. U.S. Dep’t of Defense, No. 2:25-cv-41 (E.D.N.C. Apr. 2026) (order of public reprimand), via Courthouse News.
- Thomas v. U.S. Dep’t of Justice, No. 5:25-cv-242 (E.D.N.C.), Doc. 25, Federal Defendants’ Response to Court Order (filed July 15, 2026).
- Coverage: ABA Journal (Mar. 11, 2026); Bloomberg Law; Law360; Courthouse News.
Open-source starting points
- Claude for Legal on GitHub: about 100 skills, Apache 2.0. Start in
litigation-legal. - CourtListener: free case law, and the connector used in Part 4 to verify citations.
- kylebahr.netlify.app/2000: a running tracker of lawyers misusing AI.
The thesis
- Cuffney, Ethically Implementing Generative AI in Prosecution (Northwestern University, 2025). Six career prosecutors evaluated AI summaries across ten redacted public-records cases; their criticisms became the rules in the prosecutor-analysis skill.
- Statute and instruction libraries built from ncleg.gov and the NC School of Government’s pattern jury instructions.
Let’s stay in touch.
Robert Cuffney · Legal Knowledge Engineer · Former Mecklenburg County ADA