Adi's Digital Garden

Deep dive · 04

Action maturity — how a regulated agent earns the right to act

PHP's value came from doing things for patients — booking appointments, fetching coupons, ordering products. But you can't let an agent act in healthcare until it can justify why. So I modeled maturity as two dimensions that climb together.

Two dimensions, one ladder

Agent action maturity
Explanation maturity
L1
Passive recommendationRecommend a category or specialty.
Relate to the conversationTie the suggestion to what the patient just said.
L2
Explainable deep-linkLink to the specific entity — a doctor profile, a product page.
Justify the tool callExplain the parameters — why this zip, this specialty, this query.
L3
Exact item, surfacedThe precise booking slot, product, or lab test.
Explain the algorithmWhy the ranking returned this — how the API itself prioritizes.
L4
One-click actionThe agent completes the workflow — booking, purchase, delivery.
Full transparencyEvery prior rung's justification carried through.
Maturity isn't one axis. The agent's licence to act is gated by its ability to explain — they advance in lockstep.

Explainability is the gate on action

Promote an asset's action level?
Explanation hasn't caught up
A more autonomous action would act without being able to say why → not shipped. Hold the action level.
Explanation is at the same rung
The agent can justify the parameters and the ranking → safe to promote the action.
The same principle as the eval program: in a clinical product, capability you can't explain is capability you can't trust.

The ladder, walked — finding a doctor

What the agent does
How it justifies it
L1
Recommend seeing a cardiologist.
"You reported chest pain, so we recommended cardiologists as the most relevant specialty."
L2
Link to Dr. Jane Doe's profile on Healthgrades.
"Dr. Doe practices in your zip and matches the specialty we inferred from your symptoms" — why this zip, why this specialty.
L3
Link to book Dr. Doe at Clinic X on 8/15.
"Healthgrades ranks zip above specialty; this was the top result for your zip + condition" — explaining the API's ranking, not just our parameters.
L4
Automated phone booking via VAPI — the appointment is made for you.
Carries every prior justification into the action itself.
The same asset, climbing the ladder — from a passive specialty hint to a booked appointment, with the explanation deepening at every rung.

The portfolio — where each asset stood, and where it was headed

L1
L2
L3
L4
Healthcare ProviderHealthgrades
now
target
OTC ProductsOptumStore
now
target
Online CareOptum Now telehealth
now
target
ArticlesHealthline
now
target
current level target level
A single framework let me reason about four very different integrations on one scale — and sequence the roadmap by gap, not by gut.

What "fully actioned" meant for each asset

Healthcare Provider

Automated phone booking via VAPI — the appointment is placed for the patient.

OTC Products

In-app button to purchase / add to cart and get the product delivered home.

Online Care

Take async-care actions — labs, tele-prescriptions — directly in the app.

Articles

Subscribe the patient to a condition-specific newsletter.

The eventual north star: plug into e-commerce pharmacies for doorstep medication delivery — the highest-blast-radius action, and so the last to be promoted.