The other 80%: running a law practice, not just searching it

It is tempting to build only the impressive part. AI search over twenty million judgments is the thing that gets a demo audience to lean forward. But an advocate's day is mostly not research — it's matters, clients, hearing dates, documents, drafts, invoices, and a dozen small promises made in corridors.

If those live in a diary, WhatsApp and a filing cabinet while the AI lives in a browser tab, the AI becomes a novelty. The research product only compounds if it sits inside the system where the practice already keeps its work.

And 'inside the system' means mobile. Advocates are in transit, in court, and in corridors far more than they are at a desk. A desktop-only legal platform is a platform used at 9pm, badly.

So the suite had to cover the boring half properly: matter and client management, document storage per chamber, drafting from a large template library, limitation and hearing-date tracking, time capture and billing with local tax handling, and communication that reaches clients where they actually read — which in India means WhatsApp, not email.

Multi-tenancy was the architectural spine. A chamber's data is not a folder — it's a boundary. Firms share matters internally and must never leak across; documents belong to the chamber, not the individual; and the security rules for that had to be written and reasoned about directly, not assumed.

In-app document editing was a deliberate, expensive call. Advocates live in .docx. Sending them out to another application to draft is where a workflow dies, so we self-hosted a full document editor rather than accept the round trip.

The mobile decision was the biggest one I influenced. The first app was built natively-cross-platform in one framework; the rebuild moved to React Native with Expo so that web, iOS and Android could ship from one codebase and one roadmap. That is a product decision disguised as a technical one: with a small team, three platforms means either one shared codebase or two permanently neglected platforms.

Research then stops being a destination and becomes a capability *inside* the workflow — you're in a matter, you need authority, and it's there, with the citation ready to drop into the draft you already have open.

The result is a suite rather than a feature: matters and clients, chamber-scoped documents with in-app editing, a large drafting template library, hearing and limitation tracking fed by automated court sweeps, time capture and billing, and client communication over the channel Indian practices actually use — all reachable from a phone in a corridor as easily as from a desk.

Research and case tracking then plug into that spine instead of standing beside it, which is what turns an AI feature into an operating system for a practice.

The lesson I keep from this one is about where AI belongs. An AI feature parked next to someone's real workflow is a demo; the same feature embedded inside the system where their work already lives is a product. Most of the effort that makes AI valuable is spent on the unglamorous software around it.