I asked Claude AI what artificial intelligence agents, super agents, and/or apps it recommends I start building now.
Here’s part of what it said:
A healthcare project.
The Founding Frustration
Ideation: identifying what’s wrong before imagining what’s right
The project doesn’t begin with a vision. It begins with an irritant.
The state of personalized medicine right now is philosophically dishonest. Not ineffective. Dishonest. The field calls itself personalized but operates by segmenting populations into progressively smaller groups and calling the smallest group an individual. A treatment is “personalized” when it fits your genetic cluster, your demographic cohort, your biomarker range. It is never actually you.
What would a health system look like if it started with exactly one person and never generalized outward?
This would require a model of the body that holds all of a person’s contradictions simultaneously — their stated health goals and their actual behaviors, their genetic predispositions and their emotional overrides, their rational supplement regimens and their irrational 2am eating — without resolving these contradictions into a simplified profile.
Building the Map of Existing Failures
Research phase: learning the field well enough to locate its ceiling
Rigorously study not what current health AI can do, but the precise shape of what it cannot.
Identify three structural ceilings that classical approaches cannot escape:
The snapshot problem. Every health model, however sophisticated, is trained on historical data and makes predictions about a future person who is assumed to resemble the historical average. But individual bodies are dynamic systems — a person’s optimal sleep architecture at 28 is not the same as at 22, and no model trained on population data can predict that individual trajectory without continuous recalibration.
The emotion-biology separation. Clinical health AI treats emotional and psychological data as a separate input stream — something to be correlated with physical health metrics after the fact. Grief is a variable. Chronic low-level anxiety is a variable. This is wrong: emotional states don’t correlate with biological states from outside. They are biological states, expressed in a different register. A system that processes them separately is modeling a fiction.
The irrationality discard problem. Health AI systematically filters out behavioral data that doesn’t conform to expected health-seeking patterns — the skipped medications, the abandoned exercise regimes, the dietary chaos. These are treated as noise obscuring the true signal of what the person “actually wants.” But the irrational deviations are some of the richest signal the body produces. They are the places where the person’s stated preferences and their actual biological needs are most visibly in conflict. A system that discards them is discarding the most important data.
Collaborate with a biotech researcher or a clinical data scientist who doesn’t refute this conceptually.
First Principles in Biotech
Learning the biological substrate your build will need to model
Acquire working literacy in four areas, pursued not comprehensively but strategically — always asking what does this domain know that my system will need to model?
- Genomics and epigenomics. Genomics is the study of your DNA — the complete instruction manual inside every cell that determines traits like eye color and height. Epigenomics studies the “switches” that turn those instructions on or off. Things like diet, stress, and environment control these switches — without changing the DNA itself — affecting how your genes actually behave.
- The microbiome as a complex adaptive system. Your microbiome is trillions of bacteria, fungi, and viruses living in your gut. Like a rainforest, it’s a complex adaptive system — the microbes constantly interact, compete, and cooperate, responding and reorganizing when things change (like a new food or illness). No single microbe is in charge; the whole community adapts together to keep you healthy.
- Chronobiology. Chronobiology is the study of how living things follow internal time schedules. Your body runs on a roughly 24-hour circadian rhythm — a built-in clock that controls when you feel sleepy, hungry, or alert. Light and darkness help reset this clock daily. When your schedule fights your biology — like staying up late — your health can suffer.
- Psychoneuroimmunology. Psychoneuroimmunology studies how your mind, brain, and immune system talk to each other. Stress and emotions send chemical signals that can weaken your immune defenses — that’s why you might get sick after a tough week. But positive feelings and strong social connections can actually strengthen immunity. Your thoughts and feelings are literally shaping your body’s ability to fight disease.
The First Agent Architecture
Build a prototype
Start with a first prototype agent designed to do one thing: hold a single person’s health data stream without collapsing it into recommendations. No interventions. No optimization. Just sustained, multi-modal attention to a living biological system.
You’ll likely find that the project needs quantum computing. Not for speed. For the ability to hold relationships between data streams operating at radically different timescales in superposition — to preserve the full complexity of the system before any optimization occurs.