The Claude-Powered Trial Lawyer

I’m a plaintiff’s personal injury trial lawyer in Georgia. I handle car accidents, medical malpractice, wrongful death, premises liability, product liability cases — really, anything involving catastrophic injuries. My cases involve thousands of pages of records and documents, dozens of depositions, and years of litigation before they resolve.

I don’t have a team of associates. I don’t have a dedicated paralegal on every case. What I do have is a system I built myself — using Claude.ai and Obsidian — that has fundamentally changed how I practice law.

A few weeks ago, Zack Shapiro’s article about building a “Claude-Native Law Firm” went viral. Over seven million views. His firm uses Claude to handle corporate and startup work — contracts, entity formation, investor negotiations. It resonated with a lot of people.

But I read it and thought: that’s not my world.

My world is complaints, discovery requests and responses, depositions, document and exhibit management, trial notebooks, and demand letters built on three years of documented injuries. My cases don’t close in days. They close in years. And the volume of information in a single wrongful death case can become overwhelming — particularly when it’s just one of the many cases I’m litigating at any given time.

So I built my own.

The Problem: Litigation Is an Information War

Here’s what most people outside of plaintiff’s Personal Injury (“PI” for short) work fully appreciates: the complexity is not in the legal theory. Negligence is negligence. The complexity is in managing and synthesizing large volumes of information and distilling it down to what matters so you can clearly and concisely explain to the jury (or the other side) why we win!

A single medical malpractice case might involve:

• Thousands of pages of records and documents across dozens of sources
• Depositions of the treating physician, the defendant doctor, the corporate representative, three fact witnesses, and two experts
• Discovery requests, responses, deficiency letters, and motions to compel
• A trial notebook that maps every element of every claim to specific evidence in the record
• A demand letter that synthesizes all of it into a persuasive narrative for an insurance adjuster

Now multiply that by twenty active cases. Some in discovery. Some approaching trial. Some waiting on expert opinions. Some in settlement negotiations. All of them with deadlines that don’t care whether you remember them or not (and failing to meet a deadline can have devastating consequences to your case).

That is the environment I built this system for.

The System: Obsidian + Claude

I manage my entire practice in Obsidian — a free, open-source knowledge management application that stores everything as plain-text markdown files directly on my local hard drive. Nothing is in the cloud. Nothing is locked behind a proprietary database. Every note, every case file, every task, every person, every legal authority I’ve ever researched — it’s all in one vault, cross-linked, searchable, and structured.

Claude operates inside that vault.

Not as a chatbot I ask questions to. As an integrated system with defined roles, documented skills, and structured workflows that run against my actual case files.

The AI Team

This is where it gets different from most “I use AI in my practice” stories.

I don’t use Claude as a single tool. I built a team of specialized AI personas — each with a defined role, domain expertise, documented workflows, and specific responsibilities. They are activated through Claude’s Projects and Skills features, and they operate on my actual case data (anonymized before processing — more on that below).

Here’s the team:

Lincoln is the orchestrator. When I bring a task, Lincoln identifies which team member should handle it and routes accordingly. Lincoln never executes work directly — Lincoln delegates.

Posner is my senior legal researcher. Named after Judge Posner. Posner researches legal issues using a strict source hierarchy: first the vault (my own case law notes and statutes), then Google Drive references, then web research — but web sources are never treated as authority. Every proposition gets a confidence rating. Every research session ends with the vault updated. Posner has a zero-tolerance hallucination policy — if a case can’t be verified, Posner says so and proposes a Westlaw search instead of making something up.

Atticus is my litigation strategist. Named after Atticus Finch. Atticus starts at the verdict form and works backwards: what does the jury need to find, what evidence establishes each finding, is that evidence in the record, and if not, how do we get it? Atticus builds proof maps, designs discovery strategy, writes deposition purpose statements, and identifies gaps in the record — honestly. When the case is weak on causation, Atticus says so.

Scalia is my senior drafter. Named after Justice Scalia — because whatever you think of his jurisprudence, the man could write. Scalia takes Posner’s research and Atticus’s strategy and produces court-ready documents: demand letters, complaints, discovery requests, motions, briefs, deposition outlines, client letters. Everything is delivered as a properly formatted Word document with real styles — not copy-paste from a chat window.

House is my medical research assistant. Named after the TV character. When I’m working through complex medical issues — questions of causation, standard of care, mechanism of injury — I can feed anonymized medical facts into House and get back substantive analysis. What does the medical literature say about this mechanism? What are the accepted standards for this type of treatment? What questions should I be asking the expert? What do published industry standards say about the “best practices” in the field? House doesn’t replace expert testimony. House helps me prepare to ask better questions and understand the medicine well enough to litigate it effectively.

Harvey is my settlement analyst and case valuation strategist. Harvey builds damages frameworks, researches comparable Georgia verdicts, maps insurance coverage, and advises on demand strategy. When I need to explain case value to a client, Harvey drafts the communication in plain language.

Erin is my litigation paralegal. Erin owns the document registry for every case — a master exhibit index with permanent ID numbers that track documents from intake through deposition through trial. When defense productions arrive, Erin processes them against the original requests and flags what’s missing.

Radar is my case manager. Radar tracks every deadline, every task, every client communication touchpoint across all active cases. Nothing falls through the cracks because Radar’s job is to make sure nothing falls through the cracks.

There are others. The point isn’t the number. The point is the architecture.

How It Actually Works: A Real Workflow

Let me walk through what happens when I receive defense discovery responses in a case. This is one of the most time-consuming tasks in plaintiff’s litigation, and it’s where the system earns its keep.

Step 1: The responses arrive. Erin processes them — checks page counts against the cover letter, verifies Bates ranges, scans the privilege log, sorts documents by type, and produces a production summary.

Step 2: Atticus reviews the responses against our discovery matrix — a table that maps every element of every claim to the specific discovery vehicle designed to establish it. Atticus produces a ranked deficiency analysis: what was adequate, what was improperly objected to, what’s incomplete, and what’s missing entirely.

Every deficiency is ranked by case impact: case-dispositive gaps get flagged red. Damages-significant gaps get flagged yellow. Impeachment material gets flagged green. This is not a generic “they didn’t fully respond” letter. It’s a surgical assessment of what’s missing and why it matters.

Step 3: Scalia takes Atticus’s deficiency analysis and drafts a Rule 6.4 letter — Georgia’s meet-and-confer requirement before filing a motion to compel. The letter itemizes every deficiency with the legal argument for why each objection fails.

Step 4: Radar calendars the response deadline, the motion-to-compel window, and creates follow-up tasks.

That entire workflow — which used to take me the better part of a day per case — now takes a fraction of that time. And the output is better, because nothing gets missed. The system is more thorough than I am on my best day, and it’s consistent on its worst.

Depositions: Where the System Really Shines

I built dedicated Claude Skills for different types of depositions: at-fault drivers, defense experts, fact witnesses, treating physicians, corporate representatives. Each skill is a detailed instruction file — not a template, but a methodology.

The at-fault driver deposition skill, for example, doesn’t just list questions. It encodes deposition doctrine: the five purposes of cross-examination, how to structure propositions, when to use leading questions versus open-ended ones, how to lock in admissions before attacking. It includes Georgia-specific considerations — the rules of the road, traffic code provisions, electronic data recorder preservation requirements.

When I’m preparing for a deposition, I activate Atticus to produce a purpose statement — what must this deposition accomplish, and what specific commitments need to be locked in. Then the appropriate skill generates a complete, case-specific outline with actual questions, exhibit references, and impeachment material drawn from the case file.

The outline is delivered as a formatted Word document with proper heading styles, so I can use Outline View to navigate during the deposition itself. Blue headings, italic quotes for prior statements, red bold for exhibit hand-offs. It’s built for the courtroom, not the screen.

Trial Notebooks: Building the Record from Day One

Most lawyers start thinking about trial sixty days before the date. I start building the trial notebook at case filing.

The Trial Notebook skill generates a structured litigation-ready framework: causes of action with Georgia elements, an element-by-element proof roadmap, a damages framework, anticipated defenses with rebuttal strategies, and trial themes.

Atticus’s trial strategy framework is built on jury decision-making research — specifically the Story Model (Pennington & Hastie), which shows that jurors don’t tally legal elements. They construct a narrative. A causal story that is coherent, morally meaningful, and emotionally honest. Then they match it to the verdict form.

So every trial notebook starts with four questions:

1. Who made the choice?
2. Why was it wrong?
3. What happened because of it?
4. Why does justice require this verdict?

The legal elements are mapped to evidence. The evidence is mapped to witnesses and exhibits. Gaps are identified early — while there’s still time to fill them. By the time trial approaches, the notebook isn’t a scramble. It’s a culmination of systematic preparation.

The Knowledge Compound Effect

Here’s what most people miss about this approach: it compounds.

Every deposition summary Posner creates becomes a permanent asset in the vault. Every case law note is cross-linked to legal topics, statutes, and other cases. Every deposition outline builds on the doctrine encoded in the skills. Every Rule 6.4 letter teaches Scalia something new about how a particular defense firm responds to discovery.

After a year of this, my vault contains hundreds of case law notes, dozens of deposition summaries, a growing library of legal topics, and a set of skills that get better every time they’re used. When I start a new case, I’m not starting from zero. I’m starting from everything I’ve already built.

This is the real advantage. It’s not that any single task is dramatically faster — although many are. It’s that the system learns. Not in the machine-learning sense. In the institutional knowledge sense. The vault gets smarter, and every new case benefits from every case that came before it.

Anonymization: The Non-Negotiable

I am acutely aware that I’m feeding case information into an AI system. So I built an anonymization protocol that runs before any document is processed.

Every document — court filings, depositions, correspondence — is anonymized before Claude sees it. Names become PERSON_01. Organizations become ORG_03. Case numbers, addresses, dates that could identify someone — all replaced with pseudonyms. The mapping database lives only on my local machine.

If Anthropic were subpoenaed tomorrow, their servers would contain pseudonyms with no way to map them back to real identities. This is not optional. This is how it works every time.

After Claude processes the anonymized content and produces output, the pseudonyms are restored before anything is written to the vault or delivered as a document, but not in the cloud — locally on my computer. The anonymized version exists only during processing.

What This Is Not

This is not a pitch for a product. I’m not selling software. There is no waitlist.

This is a description of what I built with tools that are available to anyone. Claude costs about $100 a month. Obsidian is free. The skills and team member files are plain-text markdown documents I wrote myself.

The barrier to doing this is not technical skill. I’m not a developer. I have never written production software. I built this system by talking to Claude about what I needed, iterating on what worked and what didn’t, and encoding the successful approaches into reusable skills. And the best part is if you have a thought or an idea and don’t know how to implement it, just ask Claude — it knows.

The barrier is imagination. And time. And the willingness to invest both into building something instead of buying something.

Why I’m Writing This

Zack Shapiro made a compelling case for the Claude-native law firm in the transactional context. But litigation is different. The information density is higher. The stakes are different — these are people who were hurt, and the quality of the work directly affects whether they get justice. The workflows are longer and more complex.

I wanted to show that this approach works here too. Maybe especially here.

I also wanted to say something to the lawyers who are watching the AI conversation from the sidelines, feeling like it’s not for them. Feeling like it’s for tech-savvy corporate lawyers, not for the trial lawyer who’s been doing this for twenty years with a yellow legal pad and a banker’s box of documents.

My dad is 81 years old. He’s a retired lawyer. He called me the other day, excited to tell me he’d discovered ChatGPT and was using it to help his neighbor with a legal problem. He went on for ten minutes about how it helped him find the right statutes, figure out where to file, and draft a letter. Then he asked me if I’d ever used AI before.

If he can figure it out at 81, you can figure it out too.

I don’t think AI is going to replace lawyers. But I do think lawyers who use AI are going to replace lawyers who don’t.

The tools are here. The question is whether you’re going to build something with them.