Technology’s Effect on Advocacy
drawer. Most firms have never built one, and no off- the-shelf product does it for you because the specific categories depend on how your firm thinks about its own work. I realized that this gap sits beneath every product on the market. The ask was never "find me an AI that drafts complaints." The ask was "build the foundation that turns my firm's body of work into something any good AI can use." Once that exists, the model on top essentially becomes a commodity. Swapping one for another comes down to token cost, not performance. A firm’s true leverage lives in the structured exemplar layer underneath, and your firm owns that layer. No one else has your work product, and no one else thinks about it the way you do. This realization sent me back to my laptop.
For the lawyers who look at what they can rent and think "there should be something more" — this is for you, and the barrier to entry sits far lower than it did just two months ago. Week One. I opened Claude Code, Anthropic's command-line coding assistant, and typed in plain English: "Build me a Word document with line numbers down the left margin from 1 to 28, locked to the page, that won't drift when I add a section break." By the end of the week I had a working pleading- paper builder without writing a single line of code by hand. I described what I wanted, looked at what came back, said what was wrong, and iterated. That foundational skill has nothing to do with coding. You use it every time you tell a junior associate what a brief should look like and then redline what they bring back. Lawyers already excel at this. Week Two. I started bundling those prompts into reusable artifacts — “skills” in the Claude ecosystem. A skill packages instructions, reference material, and example outputs so the AI loads the right context whenever a particular kind of work shows up. First the pleading-paper builder, then a letterhead generator that finally tamed the partner block. Week Three. Skills can call other skills, and chaining them creates a plugin: related skills bundled into one tool to carry an entire job from start to finish. I built a complaint drafter that knows our structural conventions, retrieves a closely-matched exemplar from our own filings, and produces a first-pass draft on conforming pleading paper. Then I indexed our body of filings into the firm's foundational library for that plug-in. The exemplar retrieval is the secret: the model writes like our firm because it reads our complaints first, not because it reads the internet. Where the library runs thin, the system researches
Six Weeks, In Honest Order
Here is what the learning curve looked like for me — written for the lawyer wondering whether they could do the same. Most of you could. The barrier is not intelligence; every trial lawyer I know is smart and discerning enough to build this. The true barrier is the false belief that this is somebody else's job. Candidly, the work costs something. I build on nights and weekends, much of it bug hunting; the firm now doubles as an early adopter, my own team reporting what breaks in real time. I keep at it because I am building an asset: nine to eighteen months out, the plan has us substantially automating intake, record retrieval, lien notice, pre-litigation workflow, most bookkeeping, and settlement distribution. All of this is in service of creating a world-class service experience for our clients, reserving more time for our staff to work up our cases and communicate with our clients. For most firms, off-the-shelf remaints the correct
67
Consumer Attorneys of California
FORUM September/October 2026
Made with FlippingBook - Online catalogs