Field Note: 21 Ways AI Can Hallucinate in Your Legal Brief

Episode 014: Billable Hours and the AI Native Law Firm

SHOW NOTES  Is the billable hour a liability you're voluntarily handing to your clients — and is AI finally giving lawyers a way out? Rich Rodgers has been building AI-native legal tools since before most lawyers knew what a large language model was. He's a practicing startup attorney, a four-time founder, and the creator of Start Legal — an AI platform designed to give founders a running start on legal work before they ever engage counsel. This episode isn't about whether AI will replace lawyers. It's about whether lawyers who refuse to adapt will replace themselves. In this episode: * How Rich Rodgers built Start Legal out of 20+ custom GPTs originally created for his own clients at Startup Tech Law * Why Rich charges $10,000 to review AI-generated client contracts — and why no one has ever taken him up on it * The "Attorney Assist" model: how Start Legal pairs AI-generated documents with on-demand attorney review starting at $150 for 30 minutes * What "AI-native law firm" actually means — and why it's harder to define than the people throwing the phrase around on LinkedIn let on * The argument for scrapping the billable hour: if AI cuts a 5-hour task to 5 minutes, does the client owe you for the hours you didn't spend? * Rich's AI-generated invoicing workflow — how he wakes up on the 1st of every month with all client invoices already built and ready to send * The Delaware ruling allowing corporations and other artificial entities to vote in certain municipal elections — and what it might mean for corporate governance documents * A concrete FSJ roadmap: what Flintstones-, Simpsons-, and Jetsons-level lawyers should actually do next We also discuss: * The em-dash problem: how a punctuation mark became a tell for AI-generated text (and Ron's personal defense of it) * Whether AI-native practice is more natural in transactional work than litigation — and where the limits actually are * The UK firm licensed to practice law as an AI — and the startup firm handling traffic tickets with no human attorney * Whether Flintstones lawyers should only try to serve Flintstones clients, or whether tech alignment between lawyer and client matters * Why Rich thinks the next generation of clients will arrive already fluent in Claude and GPT — and what that means for practices that aren't ready Key Takeaway Availability is not authority — and it's not a business model either. Clients are already arriving with AI-drafted contracts, AI-researched questions, and AI-generated documents they believe are finished products. The lawyers who treat that as a threat are the ones charging $10,000 for a GPT contract review as a way of saying no. The lawyers who treat it as an opportunity are building the tools, setting the terms, and staying in the loop on their own conditions. The Flintstones lawyer's first move isn't to become a Jetsons lawyer overnight. It's to take whatever templates, clauses, and hard-won knowledge are sitting in a file cabinet — or a Microsoft Word folder — and start turning them into something that works for clients instead of just for the file. The Simpsons lawyer who's already prompting should be connecting those prompts to the operational infrastructure: billing, CRM, invoicing. The Jetsons lawyer is already doing what Rich is doing. The question for everyone else is how long the gap keeps widening. Mentioned in This Episode * Rich Rodgers — https://www.linkedin.com/in/richrodgers360/ [https://www.linkedin.com/in/richrodgers360/] * Start Legal — https://startlegal.com/ [https://startlegal.com/] * Startup Tech Law — https://www.startuptechlaw.com/ [https://www.startuptechlaw.com/] * Claude (Anthropic) — https://claude.ai/ [https://claude.ai/] * ChatGPT / OpenAI GPT Store — https://chatgpt.com/gpts [https://chatgpt.com/gpts] info@drescherlaw.com [info@drescherlaw.com]

Workflow Options: Claude for Legal and Strongsuit

Workflow Options: From Prompts to Presets Every lawyer who has ever stared at a blank prompt box knows the feeling. AI promised to change how legal work gets done — but a single chatbox isn't a workflow. The legal AI world is splitting in two: enterprise ecosystems building choreographed plugin infrastructure, and vertical tools purpose-built for specific practice areas. The real question isn't which AI is smartest. It's which platform removes the most friction for your practice. In this episode: * Why the AI world is moving from one-off prompting to persistent, repeatable workflow environments * What connectors, APIs, MCPs, plugins, and skills actually mean — and how they differ from each other * What Claude for Legal launched: 22 connectors, 12 plugins, and MCP architecture designed for enterprise legal environments * Why Claude for Legal is largely out of reach for solo and small firm lawyers — and why that's not necessarily a problem * What StrongSuit is, how its divorce and family law workflow presets work, and what "blank prompt box anxiety" means for practicing lawyers * How Markdown (.md) workflow files work, why they're portable across frontier AI tools, and how Ron built his own We also discuss: * Why Consumer Claude Pro is not three-legged stool compliant for confidential client data * The M&A demo Anthropic ran at their Claude for Legal launch — and what it signals about their actual target market * How enterprise platforms like Harvey, Copilot, and Gemini fit into the ecosystem track * Vertical specialty tools beyond StrongSuit: Litmus.ai, Glade.ai for bankruptcy * The "blank page anxiety" finding from the big law associates episode and how presets address it Key Takeaway The AI tool that wins your practice isn't the one with the most connectors or the highest benchmark scores. It's the one that eliminates the friction between where you are and where the finished work product needs to be. Presets and persistent workflows do that in a way raw prompting never could. If you're a Simpsons lawyer — aware of AI, dabbling, maybe running isolated prompts — this episode is your map. You don't have to build enterprise infrastructure. You need to identify one workflow bottleneck in your practice and find the tool that addresses it specifically. For family law lawyers, StrongSuit may already exist. For others, a Markdown workflow file built with your AI may be closer than you think. Mentioned in This Episode * Claude for Legal [https://claude.com/blog/claude-for-the-legal-industry] — Anthropic's official announcement (22+ connectors, 12 plugins, MCP architecture) StrongSuit [https://strongsuit.com/] — divorce and family law AI workflow platform Model Context Protocol (MCP) [https://modelcontextprotocol.io/docs/getting-started/intro] — official documentation Harvey [https://harvey.ai] iManage [https://imanage.com] NetDocuments [https://www.netdocuments.com] Thomson Reuters Co-Counsel [https://legal.thomsonreuters.com/en/products/co-counsel] Westlaw [https://www.westlaw.com] Free Law Project [https://free.law] DocuSign [https://www.docusign.com] Litmas.ai [https://litmas.ai] Glade.ai [https://www.glade.ai] Gemini [https://gemini.google.com] Microsoft Copilot [https://copilot.microsoft.com] Ivory Mind [https://www.ivorymind.com/lawyers] ChatGPT / OpenAI GPTs [https://chatgpt.com/gpts] Google Gems [https://gemini.google.com/gems] Prior podcast episodes: Episode 007: Folder Mania — AI Comes to You [https://lawyeraitoolkit.com/episode-007-folder-mania-ai-comes-to-you] Field Note: I Wanna Hold Your Hand — Learning AI from AI [https://lawyeraitoolkit.com/i-wannai-hold-your-hand-learning-ai-from-ai] How BigLaw Associates Are Actually Using AI in Legal Drafting [https://lawyeraitoolkit.com/how-biglaw-associates-are-actually-using-ai-in-legal-drafting]  info@drescherlaw.com [info@drescherlaw.com]

Episode 013 BigLaw, Privilege and an Unexpected "Wow!" Moment

When AI becomes a privilege problem, most lawyers are still treating it like a productivity hack. Solo and small firm attorneys hear constantly that AI saves time. What they hear far less often is that the AI tool they chose — and more specifically, the tier they're using — may have just waived their client's privilege. This episode forces that conversation. If you're putting client material into any AI tool without understanding exactly how that tool handles your data, you're not just taking a risk — you're potentially handing opposing counsel a gift. In this episode: * Why the tier of AI tool you're using (free, Pro, Enterprise) is a privilege and confidentiality issue, not just a performance issue * The U.S. v. Heppner case: how using Claude at the Pro tier — not Enterprise — led to a court finding that confidential materials weren't protected * The Trembly v. OpenAI case (2024 U.S. Dist. Lexis 141362) and what it establishes about AI outputs as opinion work product * Why Harvey's architecture makes it suitable for confidential client material when Claude's public tiers do not * How to build privilege defensibility into your AI workflow: mandatory human review, output labeling, and policy documentation * Matt Lafferman's framework for ensuring AI outputs qualify as opinion work product rather than discoverable fact work product * Ron's "record everything" hot take — and Matt's push back on where that logic breaks down in sensitive matters * How in-house counsel faces a unique AI challenge because business and legal functions blur, and only the legal portions are privileged * Automating law firm intake: what tools like Clio, MyCase, and PracticePanther are building, and what Dentons is already doing We also discuss: * How Dentons uses Harvey as a document vault, including running tiered relevancy scoring on large document sets * The README file as the next frontier: why tech-sector in-house counsel may need to rethink document formats entirely * Ken Griffin's reversal on AI — from calling it "garbage" in January to expressing alarm at what agentic AI is doing to PhD-level work * Whether AI saves time or elevates work product quality — Ron and Matt respectfully disagree * The challenge AI poses for junior associate development and entering the legal market right now * The FSJ closing segment: Flintstones, Simpsons, and Jetsons advice from a Big Law AI task force member > Download: Notice of Intent to Use AI in Discovery Helps lawyers disclose AI use in litigation in a structured, defensible way. Covers: > * When and how to disclose AI use to opposing counsel > * Language for protective orders that includes AI tools in the definition of authorized agents > * How to frame AI outputs as generated at the direction of counsel > * Adaptable to different jurisdictions and risk profiles Key Takeaway Availability is not authority — and that principle extends to tool tiers. Using an AI tool that collects your prompts, trains on your outputs, and discloses data to third parties isn't just a privacy concern. It's a privilege waiver waiting to happen. Matt Lafferman's framework is straightforward: choose the right tool, mandate human review, mark everything as work product, and document your policy so you can show a court exactly how your AI workflow maintains privilege at every step. For Flintstones lawyers, this episode is a fire alarm — the risks are real and courts are already ruling on them. For Simpsons lawyers using Claude Pro or a free tier for anything client-adjacent, this is the moment to audit your setup. Jetsons lawyers building custom agents should be baking these privilege protections into their workflow architecture from day one, not retrofitting them after a discovery dispute. Mentioned in This Episode * Matt Lafferman, Partner, Dentons (white collar, government investigations, crypto/blockchain, AI task force) [https://www.dentons.com/en/matthew-lafferman] * Harvey (enterprise AI platform for legal) * Claude (Anthropic) — Pro tier vs. Enterprise tier distinction * Microsoft Copilot * Clio * MyCase * PracticePanther * U.S. v. Heppner — Claude Pro tier, privilege, and work product * Trembly v. OpenAI, 2024 U.S. Dist. Lexis 141362 — AI outputs as opinion work product * Ken Griffin / Citadel — remarks at Stanford Business School on agentic AI * Judge Rakoff, Southern District of New York * Rule 26(f) AI Discovery Protocol Addendum * Notice of Intent to Use AI in Discovery * README / .md files as emerging document format for in-house counsel * WhatsApp communications in crypto litigation * Daubert motions * Ron's "record everything" CLE hot take > info@drescherlaw.com [info@drescherlaw.com]

Field Note: I WannAI Hold Your Hand: Learning AI From AI

The AI training market for lawyers is broken. Not because there isn't enough of it — there's more than ever. The problem is almost all of it is aimed at the wrong lawyer. LinkedIn is full of Jetsons lawyers talking to other Jetsons lawyers, while the majority of practitioners are still trying to figure out how to create a PDF. So if the training doesn't meet you where you are, what do you actually do? In this episode: * Why the explosion of AI training has created more confusion, not less, for solo and small firm lawyers * The "screenshot, upload, prompt, repeat" method Ron uses to navigate new software with AI as his guide * Why AI has become an always-on tutor — and where that tutor reliably falls short * The maze-solving metaphor: how AI-guided learning gets you there eventually, but not always efficiently * Three diagnostic questions to ask any AI trainer before you spend a dollar or an hour * Why governance is no longer a procedural afterthought — it's substantive, and good training has to address it * Two new downloadable deliverables: a trainer vetting checklist and a due diligence inquiry template We also discuss: * Why Microsoft and Google already have excellent AI training most lawyers don't know they have access to — Microsoft Copilot legal training (Microsoft Learn) [https://learn.microsoft.com/en-us/training/modules/empower-workforce-microsoft-365-copilot-legal-use-case/] · Google AI Skills Hub [https://ai.google/learn-ai-skills/] · Google Workspace AI for Legal [https://workspace.google.com/resources/ai-for-legal/] * The real value of human trainers: shortcuts, judgment, accountability — things AI can't reliably replicate * A real example of a law firm that made AI training stick through scheduled, team-wide calendar commitment * Ron's earlier Field Note on taking screenshots — and why, in retrospect, that wasn't absurdly basic at all — Episode 003: Specialist AI Tools [https://lawyeraitoolkit.com/episode-003-ai-specialists] * The hallucination field note: 21 ways AI Hallucinates in your Legal Brief — download at lawyeraitoolkit.com/deliverables [https://lawyeraitoolkit.com/deliverables] > Download: AI Trainer Vetting Checklist + Inquiry Template Two tools to help lawyers evaluate prospective AI trainers before investing time, money, and trust. Available free at lawyeraitoolkit.com/deliverables [https://lawyeraitoolkit.com/deliverables]. > * Three diagnostic questions to assess whether a training program fits your FSJ level > * Whether training is workflow-focused or just a feature parade > * How to assess whether the program takes hallucinations, confidentiality, and governance seriously > * A ready-to-send email/DM template for due diligence outreach to prospective trainers Key Takeaway AI is an infinitely patient tutor. It will walk you through the maze, one wall at a time, and it will eventually get you there. But it won't always get you there efficiently, and it won't always get you there correctly — especially when its training data is three years out of date. The real skill is knowing when to use the bot and when to find the human who can point you at the exit in thirty seconds. This episode speaks directly to Simpsons lawyers who are doing what Simpsons lawyers do: picking up AI tools, bumping into walls, and figuring it out one screenshot at a time. But Flintstones lawyers who haven't entered a single prompt yet will find the framework here — especially the three questions — genuinely useful before they spend anything. And Jetsons lawyers building agent workflows have likely already internalized everything Ron says. This one isn't for them. Mentioned in This Episode * Claude (Anthropic) [https://claude.ai] * ChatGPT (OpenAI) [https://chatgpt.com] * Google Gemini [https://gemini.google.com] * Microsoft Copilot [https://copilot.microsoft.com] * GoHighLevel [https://www.gohighlevel.com] * Cogent Marketing * Google Drive · Microsoft Word · Zoom * Heather Gardner (co-host) * FSJ Framework (Flintstones / Simpsons / Jetsons) * Ron's Field Note: Confessions of an AI Hallucinator [https://lawyeraitoolkit.com/confessions-of-an-ai-hallucinator] * 21 Ways AI Can Hallucinate in Your Legal Brief — download [https://lawyeraitoolkit.com/deliverables] * lawyeraitoolkit.com/deliverables [https://lawyeraitoolkit.com/deliverables] > info@drescherlaw.com [info@drescherlaw.com]

Episode 012 – AI CLEs, Flintstones Lawyers & the Problem With Legal AI Training

I recently participated in a live AI panel at the Maryland Bankruptcy Bar Association Spring Break Weekend — one of the major annual CLE and networking events for Maryland bankruptcy lawyers. The panel featured retired federal judge Paul Grimm as moderator, along with Patti Jefferson, Nancy Rapoport, and Ron Drescher. But this episode is not simply a replay or recap of the panel. Instead, we use the experience to explore a much bigger question: Is the legal profession actually teaching AI effectively? In this episode: * Why AI CLE panels may struggle to teach lawyers at vastly different technology levels * The continuing evolution of the Flintstones / Simpsons / Jetsons framework * Why some lawyers may never need to become “Jetsons-level” AI users * Patti Jefferson’s live AI demonstration using the Red Lobster bankruptcy confirmation order * The rise of agents and workflow automation beyond traditional prompting * Nancy Rapoport’s warnings about hallucinations, supervision, and professional responsibility * Why “hallucination verification” may erase much of AI’s promised time savings * A comparison between AI hallucinations in law versus medicine * Why many firms remain stuck at “sub-Flintstones” technology levels * How to identify the pain points inside your law firm before adopting AI * Why your firm should conduct a “Tech Stack Audit” We also discuss: * Dropbox and document organization for high-volume bankruptcy practices * The growing divide between consumer AI and enterprise AI * Why “ChatGPT” is no longer a meaningful description without understanding the underlying plan and governance structure * The importance of the “Three-Legged Stool” for safe legal AI deployment: * Vendor protections * Proper configuration * Human supervision Download: Tech Stack Audit Spreadsheet [https://assets.cdn.filesafe.space/LGYwVdJSqtZFXymEmecb/media/6a0643d3a97d962a35e31188.xlsx] This episode includes a downloadable spreadsheet template designed to help lawyers: * identify all software subscriptions, * calculate monthly and annual costs, * evaluate what they are actually getting from their tech stack, * and determine where they fall on the Flintstones / Simpsons / Jetsons spectrum. If you complete the spreadsheet and would like us to discuss it anonymously (or publicly) on a future episode, send it to: info@drescherlaw.com [info@drescherlaw.com] We’d love to see how lawyers are actually building — or struggling to build — their AI and technology infrastructure. Mentioned in This Episode * ChatGPT * Claude * Gemini * Harvey * IvoryMind * Google Workspace * Dropbox * Clio * Foundation AI Key Takeaway Most legal AI education still treats lawyers as if they are all at the same level of technological fluency. But Flintstones lawyers, Simpsons lawyers, and Jetsons lawyers may not even be attending the same CLE — even if they are sitting in the same ballroom.