AI's Hidden Cost in Distributed Care

I grew up in New Delhi.

As a kid, traffic felt chaotic but legible. You could stand at a roundabout and understand how cars moved. It was messy, but it was two dimensional.

When I go back now, the city feels different. Overpasses stack on top of older roads. Elevated corridors cut across markets that once sat at grade. Traffic no longer moves on a plane. It moves in layers.

Each overpass made sense when it was built. There was congestion. There was urgency. And each one helped at that node, in that moment.

What no one measured was the coordination cost of the whole map.

This is the trap I'm seeing care organizations are walking into right now.

The Speed Seduction

Narrow AI is fast. That's its defining feature and its defining danger.

There is friction in intake. A voice agent is live in weeks. Friction in scheduling. A matching bot ships in a sprint. Friction in documentation. An AI scribe is on in a day. Each deployment has a clear before-and-after. Each has a champion, a metric, and a success story. Each feels like progress because, locally, it is.

This speed is not an accident. The market incentives currently reward speed over system design. Point solutions close faster, implement faster, and show ROI faster than platforms.

But almost no one is measuring what happens system-wide.

Three nodes: 3 edges. Ten nodes: 45. Fifteen nodes: 105.

Every new tool increases the number of coordination edges across the system. Growth is quadratic. Deployment feels linear. That gap is where complexity hides.

The Cost of Speed

Every handoff between narrow AI tools is a place where context collapses.

The scheduling bot fills a shift without seeing the credential expiration in the compliance system. The compliance system flags risk without knowing what's at stake in the active episode. The documentation AI captures the note without context on what it means downstream.

Humans re-enter to reconcile the seams. Why was this staffed? Why is this patient blocked? Why was this claim denied? Every reconciliation becomes a rule. Every rule needs an owner.

Over time, the organization becomes fast in pockets and slow in aggregate. It is navigable only by veterans - the people who hold the connective tissue between all these nodes in their heads. When they leave, which they do, the system doesn't just lose their knowledge. It loses its ability to function at the seams.

The AI that was supposed to reduce operational drag can quietly become the new source of it.

Just distributed differently, and harder to see.

An Alternative Question

Reducing edges is harder than adding nodes. That's why almost nobody does it first.

The question before deploying AI should not be: “Can this tool solve this problem?”

It should be: “What coordination burden does solving it this way introduce elsewhere?”

In distributed care, AI cannot function as a collection of utilities. It has to own an operational surface end to end.

Scheduling cannot reason correctly without availability, compliance, continuity, and overtime in the same system. Intake cannot stop at referral capture if the real risk is in getting to first care delivery cleanly. Documentation cannot operate in isolation from reimbursement logic.

Systems must be internally complex and externally simple. One accountable surface instead of seven cooperating ones. That does not mean one giant, bloated software system. It means one reasoning layer across the operational surface.

At Arya, we have made a deliberate architectural choice: reduce edges, not multiply them. That means slower specs, harder demos, and fewer ribbon-cutting moments. It also means fewer seams for operators to hold together manually.

The Real Race

The pressure to deploy fast is real. Margins are tight. Labor is constrained. Every quarter without visible AI progress feels like falling behind.

But speed of deployment is not the same as speed of improvement. Local wins can hide systemic fragility.

The organizations that will pull away are not the ones that accumulated the most AI. They are the ones that reduced the coordination required to deliver care reliably.

Overpasses compound complexity. System design compounds capability.

In the next five years, the winners in distributed care will not be the fastest adopters. They will be the best architects.

Kunal Sarda
Founder, Arya
https://www.linkedin.com/in/kunalsarda/
540-250-2633