From SOPs to agents in 12 weeks — the Hope Hospital story

Hope Hospital is a multi-specialty hospital in Nagpur, India. In February 2026, the leadership had the same question every traditional business has: where do we even start with AI? Twelve weeks later, a production agent was reading the live medicine_master table, flagging low-stock items, and dropping a structured reorder suggestion into the pharmacy team's Slack channel.

This is the actual timeline.

Weeks 1–2 · Learn

Two Champions — a senior pharmacist and the IT lead — installed Claude Code. For two weeks, every email draft, every internal memo, every report ran through the tool. Conviction built. Mediocre system-design decisions ruled out by the time we got to Wire.

Weeks 3–5 · Wire

The vault was set up — an Obsidian-backed folder, synced to a private GitHub repo, indexed into a Supabase project. SOPs that had lived in one pharmacist's head for fifteen years were written down for the first time. Slack workspace was wired. The auto-sync LaunchAgent on the lead pharmacist's Mac pushed every edit to GitHub within ten minutes; a GitHub Action mirrored it into Supabase Storage another two minutes after that.

Weeks 6–9 · Automate

First agent target: pharmacy reorder. We probed Adamrit's schema, confirmed medicine_master was the right table (not the originally guessed medication — a 30-second discovery that would have been a week-long debugging loop without it), ran an idempotent bootstrap SQL adding stock columns and a low-stock view, then shipped a Deno Edge Function on Supabase that scanned the view every six hours and posted to #pharmacy.

We left it in DRY_RUN=true for two weeks. The team got the alerts but the agent didn't actually place orders. Every alert was reviewed; we tuned the reorder-level thresholds; we caught two false positives. Then we flipped the flag.

Weeks 10–12 · Scale

Three more agents in the pipeline, each one a folder under supabase/functions/ reusing the same _shared/ Slack and types helpers. The AI Leverage Ratio dashboard went live in Week 11.

What it cost

Roughly $4k for the Discovery Sprint, and ~$60k all-in for the first year (Bettroi engineering time + API costs + Supabase + Slack Pro). The pharmacy team estimates 12+ hours/week reclaimed across the reorder process alone. The agent paid for itself in week 14.

What you can copy

The exact bootstrap SQL, Edge Function template, and Slack wiring are shipped as the Pharmacy Reorder template. The 10-step path from Hope Hospital's "where do we start" to their first live agent is the Be AI-First Playbook itself.