Medtech on autopilot: one flywheel from the manufacturer to the hospital bed.

This week, something small happened that points at something large: at its first implementation demo, Imedex became the first Healtek client connected to Claude through the MCP Connector. A distributor's operation — its accounts, devices, service history — became something you can talk to. Months of data preparation collapsed into hours.

Take that one step and extend it to its conclusion. Not one distributor connected to one AI assistant — but the whole medtech supply chain, manufacturer to distributor to hospital, running as a single, AI-driven flywheel. That's the future we're building toward, and this article is an invitation to argue with us about it.

The chain today: three businesses, three dark rooms

A manufacturer builds an outstanding device and loses sight of it the moment it ships. A distributor places it, services it, supplies it — out of an ERP, a CRM, some spreadsheets and one indispensable person's memory. A hospital depends on it daily and has no simple way to say "it's beeping" or "we're out of consumables" other than a phone call that starts a clarification loop. Every hand-off between the three is a re-keying of information the previous party already had. The chain works — medtech people are heroically good at making it work — but it works on manual effort that scales linearly with every device placed.

The flywheel: three motions, one loop

Now picture the same chain as one connected system. It spins through three motions, each feeding the next:

1 — Technology sold and placed. The sale isn't a row in a CRM; it's the birth of a device record — a passport with a serial number, a site, a PM protocol, a consumables profile — visible to the manufacturer that built it and the distributor that placed it, from day one.

2 — Service that prevents failures instead of reacting to them. Because every device carries its history, service stops being a queue of surprises. PM comes due and schedules itself. Patterns across the installed base surface before they become downtime — the goal isn't faster repair, it's a failure that never happens, and a clinician who never notices anything was at risk.

3 — Consumables that arrive before anyone asks. Usage implies demand. A placed device with a known consumption profile means resupply can be timely by default — the hospital never runs out mid-procedure, the distributor never scrambles, the manufacturer sees real consumption instead of quarterly guesses.

And then it loops: consumption and service data reveal where the next placement, upgrade or renewal should happen — feeding motion one. Each turn of the wheel makes the next turn easier. That's not three departments; that's one flywheel.

Why this becomes possible now

None of this vision is new — every medtech operator has sketched some version of it on a whiteboard. What's new is that the connective tissue finally exists. When an operation runs on a medtech-native operating layer, and that layer speaks MCP, an AI agent can work across the whole loop: read the installed base, schedule the PM, draft the resupply, flag the account gone quiet. What we saw in the Imedex demo — years of scattered data standing up in hours — is the entry ramp. The flywheel is where the road goes.

One discipline makes it trustworthy enough for healthcare: deterministic, real-data-only operation. An autopilot that guesses is worse than no autopilot. Every action in the loop must trace to a real record — a real serial number, a real service event, a real consumption figure — with gaps flagged honestly, not filled with plausible fiction.

What's left for the humans? The best part.

Here's the point of it all. When the operational loop runs on autopilot, a medtech business gets to be what it set out to be: people who bring the best high technology to hospitals and doctors. The demo in the clinic instead of the data entry after it. The conversation about what the department needs next year instead of the hunt for last year's service report. Relationships, clinical insight, judgement — the things no flywheel replaces, and the things there's never been enough time for.

An open invitation

We're building this in the open — Imedex's implementation is documented publicly as it happens. If you're a medtech distributor or manufacturer, we want your version of this argument: where does the flywheel break in your world? What would you never put on autopilot? Which motion — sales, service, consumables — would you connect first? Bring your operation and let's argue about the future →

Owner: Jan Safka · Healtek field notes, July 2026. The Imedex demo and Claude-MCP connection are real events (see journey update #1); the fully connected manufacturer-to-hospital flywheel is the direction we're building, described as vision.