Technology’s Effect on Advocacy
more broadly, routing each step to whichever AI model fits the task, weighing capability against cost.
Week Five. I stopped typing my prompts. A voice- dictation tool called Whispr lets me talk to the AI while walking the dog, driving back from a deposition, or — most importantly — sitting at my desk after a double-tap of the " fn " key. The bottleneck moved from my fingers to the quality of my thinking, exactly where it belongs. Week Six. The AI now maintains its own backlog. I describe a problem out loud — "the deposition notice skill doesn't handle a § 2020.310 subpoena differently from a notice to a party" — and a separate agent files it as a task and surfaces it when the current build is clean. I am operating something closer to a one-person product organization than a solo developer. I am the product manager.
Week Four. I started building agents, which are programs that don't wait for a prompt, but run on a schedule and bring me work that needs my judgment. The first scans every active matter each Monday morning and surfaces facts missing from our case management system, citing where in the file the answer lives. This matters because documents are snapshots: a March demand letter tells you how the firm viewed the case at one point in time, not whether the lien got reduced last week or whether the statute runs in 47 days. An agent earns trust only when each case's current truth lives in clean, structured fields. Otherwise, the AI is guessing, and guessing on a caseload is how malpractice happens.
The AI does the building. I am still, always, the lawyer.
68
Consumer Attorneys of California
FORUM September/October 2026
Made with FlippingBook - Online catalogs