Piloted by intent: the AGE, or the end of artisanal engineering

There is a sentence you hear in every healthcare IT department: “integration takes eighteen months.” Eighteen months to connect a single site. Teams of scarce experts, endless acceptance cycles, know-how that lives in consultants’ heads — and leaves with them.

This model is not slow by accident. It is slow by construction: every deployment is an artisanal prototype, rebuilt by hand, site after site.

From task to intent

At MedicalCity, the layer that breaks this model is called the AGE — Artificial General Engineer. It is not one more agent in a catalogue: it is the engineering layer that sits above all the others, and it is not piloted by tasks but by intent.

The difference is not semantic. Piloting by task means writing a ticket — “build connector X,” “configure tenant Y” — and waiting for a human to chain the steps. Piloting by intent means stating the business need:

“Deploy a site for a 300-bed facility.”

That single instruction triggers the full chain — specification, architecture, code, tests, deployment, documentation — in one supervised pipeline. What used to mobilise a team of integration experts becomes a tooled, traceable, repeatable execution.

Six strokes, like an engine

The AGE executes the engineering cycle end to end, in six strokes:

1. Understand — the business need becomes a specification.
2. Design — architecture and trade-offs, made explicit.
3. Build — code, infrastructure, CI/CD, data.
4. Verify — tests, security, compliance.
5. Operate — deployment, monitoring, incident resolution.
6. Capitalize — production feedback becomes patterns again.

The sixth stroke is the one healthcare software has never had. It is what changes the economics of the model.

Guardrails are not optional

A system that writes and deploys software in healthcare has no right to exist unguarded. Three rules are non-negotiable at MedicalCity:

• HITL by criticality: no critical flow ever passes without human validation. AI prepares, humans decide — the rule holds for engineering as it does for care.
• Independent verifier: the AGE never self-validates. A distinct verifier agent checks what the AGE produces — a separation of powers, literally.
• Audit and traceability: every artifact is versioned, every decision documented. GDPR, HDS-ready hosting, the AI Act: compliance is not a document appended at the end, it is a by-product of the pipeline.

The registry: an asset that appreciates

Every deployment feeds a skill registry: mappings, connectors, compliance templates — production-validated, versioned, audited. The AGE reuses them on the next deployment.

The consequence is arithmetic: the nth site costs a fraction of the first. Where the artisanal model keeps its know-how in consultants’ heads, ours lives in a software asset that appreciates with every production release. Code can be copied. A registry of patterns validated in production, inside the European regulatory context, cannot.

Why it is the core of RaaS

Our Robot-as-a-Service offering rests on a simple promise: a guaranteed service, at a sustainable monthly rent, largely covered by the French APA. That promise only holds economically if the cost of deployment declines site after site — robot configuration, facility connection, compliance file production.

That is exactly what the AGE does. It is not a technology demo sitting next to the product: it is the industrial engine that makes the rental model viable. And it already orchestrates the agents of our AI Factory — Specification, Configuration, Documentation, RAG — the specialised workforce that produces every artifact.

Artisanal engineering has had its time. Care does not wait.