I was a retail pharmacist for 5 years. Independent pharmacies are getting squeezed by PBMs and thin margins, and owners drown in manual work: chasing prior authorizations, re-checking claims for underpayments, calling patients who never picked up. Most of it is repetitive and rules-based, so I automate it. What gets installed: - A Command Center — a local dashboard staff open: today's fills, aged will-call, margin alerts, PA queue. - The AIOS underneath — scheduled automations that ingest the pharmacy's data, run the tools (margin recovery, PA autopilot, patient win-back), and push reports. Nobody has to click anything. The Hardware: Each pharmacy gets a Mac Mini in the back office. I set it up at home, then drop it in. It runs the automations on a schedule and holds all data + dashboards locally, encrypted (FileVault). BestRx is the first pharmacy management system we're working with (Windows-only, can't run on a Mac), so the two machines stay separate. The only thing that crosses is an exported CSV of prescription/claims data: manual export to a watched folder on day one, then BestRx's built-in Scheduled Reports auto-delivers it. BestRx also has an API integration program we're pursuing, so the plan is to move off the manual drag-and-drop CSV entirely and pull the data directly. And before that, if the Playwright CLI can drive BestRx's export screen, we'll try that first to kill the manual step with zero dependency on their side. I run and update every box remotely over Tailscale (WireGuard VPN) — SSH in, deploy a new automation, no site visit. Locked down so only my device reaches any box. Where Claude runs. Claude can't run on the Mini (cloud-only), so I call it through AWS Bedrock under a BAA, billed per token (~$30–150/mo per pharmacy). Two BAAs gate everything: pharmacy↔me, and me↔AWS (so PHI can legally touch Bedrock). The consumer Claude plan is excluded from BAAs, so it never touches patient data. How I stay HIPAA compliant: - PHI stays on the box. The only PHI that leaves goes to Bedrock, legal under the AWS BAA. - Tailscale carries me and my code, never bulk patient files. WireGuard is end-to-end encrypted; Tailscale can't read the stream (the HIPAA "conduit" position). Fallback if needed: self-hosted Headscale. - Human-in-the-loop. The AI reads docs, auto-fills forms, and flags issues. It never submits. A prior auth still needs pharmacist review and prescriber attestation under their NPI. The tool prepares, a licensed human decides. - Claude builds and tests on synthetic data; only fixed, tested code runs on real PHI. No live AI improvising on patient data in production. - Plus audit log, encrypted off-site backups, retention floors.

I've built an AI chat receptionist, handles customer messages, answers questions, books appointments, the whole thing. Works great on my end. But here's where I'm stuck: when it comes to actually connecting it to a real client's WhatsApp or Instagram account, how does that handoff work in practice? Do you ask them to share API keys? Log into their account for them? Set everything up while they watch? I can't imagine telling a local business owner "send me your Meta Business credentials" and expecting that to go smoothly. How are you guys handling this? Is there a clean way to onboard a client without it turning into a security nightmare or a trust issue?

I am planning/started developing using Claude Code. Fairly simple: The user will log onto the app I am developing, view data on a Supabase database, update and save. So far, everything is easy. Help needed:- Only the data fields to be edited will be pulled from an existing database of a web CRM system, and once the data is edited by the user using the app, the updated data must be pushed back to the CRM database again. In priciple I understand what an API must do and, at a high level, how it works, but since I have never done an API Post and Call (I am not a developer), I would like to know what I need from the developers of the CRM system If someone has a playbook info for me I can follow for that integration, I would really appreciate some help/pointers This is all I have for now, hence I would like to know what else, if I need, I should ask for: https://www.hireandservice.com/terms/developer

Honest question for the solo operators here, because I don't have a clean answer. What makes a solo operation fragile isn't the building, it's that I am the bottleneck on everything that matters. Every deliverable, every fix, every judgment call routes through me. That caps how much I can take on, and nothing ships when I step away. The obvious answer is delegate, to people or to agents. The part nobody tells you is that quality usually drops the moment you hand something off, so you end up redoing it, which is slower than just doing it yourself. So you pull it back, and you're the bottleneck again. What I've tried so far: writing the boring repeatable parts down as steps, handing off the pieces where wrong is cheap and easy to catch, and keeping a human gate on anything client facing. It helps at the edges, but I'm still the single point of failure on the core work. So the real question for anyone who's actually done it: how did you take yourself out of the critical path without the quality dropping? Was it better SOPs, hiring a specific role first, leaning on agents for a slice of it, something else? What was the first thing you successfully stopped doing yourself?

Something I keep hitting and don't have a clean answer for. Catching when an automation BREAKS is easy. It errors, nothing comes out, you get an alert. Catching when it quietly gets WORSE is the hard one. The thing runs fine, returns something that looks right, but the quality slipped and nobody notices for two weeks. For the mechanical parts (did the row get created, did the email send) this is simple. For anything open ended (a draft, a summary, a reply, a piece of content) I have no clean way to score it automatically. "It produced text" is not the same as "it produced good text." What I do now is a mix: a few hard checks on the mechanical parts, a human spot checking a sample, and saving the bad outputs so I can see patterns. It works, but it's manual and it doesn't scale past a handful of automations. So the open question for people running this in production: how are you measuring whether an agent's output is actually good over time, not just that it ran? Anyone using a model to grade another model's output, and does that actually catch the slips or does it just rubber stamp them? Curious what's working.