How I Use AI as My Personal Health R&D Department

I'm sitting at my desk, staring at a Quest Diagnostics order form, and I have this familiar feeling. You know the one—where you're about to participate in a ritual that feels simultaneously important and completely arbitrary. Annual blood work. The moment where you dutifully fast for 12 hours, get stabbed by a phlebotomist, and then receive a PDF three days later that your doctor will glance at for roughly 47 seconds before declaring you "fine."

This year, I decided to do something different. I opened a chat with my AI and said: "Let's design a lab panel that actually fits my genetics and neurotype."

What followed was a three-hour conversation that turned a routine physical into the next iteration of what I've started calling my "health operating system." But to understand why that conversation was different from every generic AI health chat you've seen, I need to explain something first.

What I Mean by "Life Model"

I'm 44, neurocomplex (autistic, ADHD, gifted—the trifecta that makes dinner parties exhausting), and I've spent the better part of three years building what I call a "life model" with AI. I'm also building an AI-powered life management app called AIs & Shine, so yes, I'm professionally invested in this idea. But I started doing it for myself long before I started building for others.

A life model is exactly what it sounds like: a persistent, structured representation of me that AI can reference across conversations. It's not just a list of facts. It's an integrated understanding that includes my genetic reports, my neurotype and psychological profile, my goals, my supplement lists, my lab history, my therapy notes, my dietary patterns, and—crucially—the themes that connect all of these.

A year ago, I got my whole genome sequenced through Sequencing.com. Not the $99 consumer version that captures less than 1% of your genome—the full thing. About 200 gigabytes of data representing every genetic variant I carry. Before AI came along, that file was essentially unusable. Now it's one of the richest inputs into my life model.

Over many months, I've fed AI enough context that it doesn't just remember facts; it recognizes patterns:

• "Jon is sensitive to methyl donors—big doses of methyl-B12 spike his anxiety."
• "Jon's gut, dopamine, and inflammation are tightly linked through his genetics."
• "Jon uses AI as external scaffolding for his ADHD architecture—he needs frameworks, not information dumps."

This matters because a generic health AI will give you generic health answers. A life-model-aware AI can say: "Given your FUT2 status plus your B-vitamin genetics plus your histamine sensitivity, here's why your B12 risk is different from average—and here's why I'm suggesting hydroxy-B12 and DAO enzyme rather than whatever random multivitamin advice you'd find on Reddit."

It's the difference between having a very smart research assistant and having a very smart research assistant who has read your entire medical file, attended your therapy sessions, and understands why you have 700 browser tabs open.

Case Study #1: The FUT2 Non-Secretor Deep Dive

A few months ago, while reviewing my genome data, I noticed a variant I hadn't explored: FUT2 rs601338 AA—what geneticists call "non-secretor" status. For most people, this is a bland label in a genetic report that gets ignored. For me, it became a design problem.

I asked my AI: "I have an inactive FUT2 secretor gene. What does that mean, and how does it impact me?"

The AI's response reframed my entire understanding. FUT2 is fucosyltransferase-2, an enzyme that puts a sugar (fucose) onto certain molecules, creating ABO blood group antigens in your saliva and gut mucus. Being a non-secretor means I don't broadcast my blood type on my mucosal surfaces. The AI put it simply: "Your blood still has a team jersey, but your gut mucus is just wearing plain clothes."

This sounds arcane, but it creates a cascade of downstream effects.

Microbiome architecture: The bacteria in your gut partially depend on those mucus sugars as food. Non-secretors have different Bifidobacteria profiles and potentially higher dysbiosis risk. Certain beneficial strains that thrive on fucosylated sugars don't get their buffet in my gut. The AI noted that without intervention, probiotics for me are just "tourists that won't stay."

Infection patterns: Non-secretors are more resistant to some norovirus strains (the "cruise ship stomach bug"—small win), but other infection risks shift in ways researchers are still mapping.

B12 absorption: This is where it got interesting. FUT2 affects intrinsic factor and the mucosal environment, which means non-secretors have statistically higher rates of functional B12 issues—even if serum B12 looks "normal" on a standard lab panel.

Inflammatory associations: Slight statistical correlations with things like Crohn's disease. Not destiny, but risk levers worth understanding.

None of this was news to medical literature. What made it useful was what came next: the AI cross-referenced my FUT2 status with my known MTHFR C677T AA variant (which reduces folate processing by ~70%), my COMT quirks, my histamine sensitivity via AOC1 variants, and my existing supplement stack. It didn't just explain FUT2 in isolation. It told me what FUT2 meant for me.

"You basically have a firmware variant," it summarized, "where your mucosal surfaces are less sugary, more particular about microbes, and slightly stingy with B12—which, layered on top of your other genetics, means your gut, microbiome, and B-vitamin status are key knobs for how good your brain and body feel on any given week."

Firmware variant. That framing stuck with me.

The "Care and Feeding" Protocol

The AI generated what it called "Care and Feeding Instructions for Jon's Gut & B12 System (FUT2 Non-Secretor Edition)." Here's the architecture:

Goals: Stable, low-inflammation gut environment. Robust B12 and methylation support that doesn't spike anxiety. A microbiome curated for my specific "firmware."

Daily food pattern: Mediterranean-ish, low junk, with consistent prebiotic fibers and resistant starch. These aren't vague "eat healthy" suggestions—they're inputs designed for my gut's specific vulnerabilities.

Microbiome strategy: Small, daily prebiotic doses (partially hydrolyzed guar gum, inulin, resistant starch from cooled potatoes or rice). Bifidobacteria-focused probiotics, specifically strains like B. infantis, B. longum, and B. breve that non-secretors often lack.

B12/methylation support: Hydroxy-B12 instead of methyl-B12 (because my anxiety profile doesn't tolerate aggressive methylation), plus riboflavin to stabilize my broken MTHFR enzyme, P-5-P for neurotransmitter synthesis, magnesium, TMG to fuel my backup methylation pathway, creatine to spare methyl groups, and glycine to buffer the whole system.

Human milk-like inputs: This was the unexpected part. The AI explained that human milk oligosaccharides (HMOs), specifically 2'-Fucosyllactose, essentially let me "run an emulator for the secretor software I wasn't born with." They feed the Bifidobacteria that my body doesn't naturally support. It's compensating for absent firmware by installing a software patch.

Case Study #2: Mapping My Supplement Pile into an Architecture

I maintain a "Personalized Supplement Strategy Card"—a detailed document listing everything I take, with doses, timing, rationale, and what I call a "Genetic Multiplier" score. The multiplier shows how much more important each supplement is for me compared to an average person: 1x is standard, 10x is a critical genetic patch.

I also track all of this in a database on jonmick.ai, so I can query my stack programmatically. (Yes, I built infrastructure for my supplements. This is what happens when a product manager gets ADHD and a genome sequence.)

I uploaded the strategy card and asked: "How do my current supplements fit into this FUT2/B12/gut framing?"

The AI read through my list and built a systems diagram. Instead of seeing 30+ supplements as independent interventions, it organized them into functional categories:

What changed after seeing this map? I could finally see which supplements were truly core versus optional. I spotted redundancy. I understood which additions (HMOs, Bifido emphasis, quercetin) were high-leverage for my specific firmware. The pile wasn't chaos anymore—it was architecture with clear priorities.

Case Study #3: Designing a Personalized Annual Lab Panel

Back to that Quest order form. I asked my AI: "I'm due for annual blood work. Help me determine the ideal labs I should request. Stack-rank them so I know what's essential versus optional."

The AI didn't just dump a list. It considered constraints: insurance realities, what primary care doctors see as "standard," and my specific risk profile. Then it generated a prioritized Lab Request Sheet with three tiers:

The AI even wrote a script for presenting this to my doctor:

"I have known genetic variants affecting B-vitamin metabolism (MTHFR, etc.), FUT2 non-secretor status (higher B12 and gut sensitivity risk), and histamine sensitivity. I'm aiming for a preventive, data-driven checkup. Priority 1 items are essential for my risk profile. Priority 2 and 3 are nice-to-have if medically appropriate; I'm willing to self-pay for some if insurance doesn't cover them."

This is the part that changed my relationship with annual labs. Instead of vibing in the dark with whatever my doctor happened to order, I now have a repeatable template. I can connect changes in my supplement protocol to objective markers. I can iterate.

How the Personalization Actually Works

Let me pull back the curtain on what makes this different from typing questions into ChatGPT.

The AI doesn't see my questions in isolation. It sees layers of context: my genetic and medical data, my psychological profile and preferences, my prior conversations (what worked, what I reacted badly to, where my doctor drew boundaries), my stated goals and constraints.

It recognizes patterns across these layers. FUT2 + MTHFR + COMT + histamine isn't four separate genetic facts—it's one coherent theme: gut barrier and microbiome → inflammation → neurotransmitters → how I feel and function. The lab choices, supplement mapping, and protocols all derive from that integrated picture.

It speaks in my language. The AI learned I need frameworks, not information dumps. It gives me "genetic multipliers" and "firmware patches" because those metaphors match how my brain processes complexity. It structures outputs in tiers because I'll actually use a prioritized list; I'll ignore a wall of text.

And it respects constraints. The AI consistently avoids prescribing or dosing like a doctor. It frames supplements and labs as conversation starters with my clinician. When it suggested L-glutamine, it flagged the anxiety risk. Safety rails, not recklessness.

How You Can Build a "Mini Life Model" for Health

You don't need to go as deep as I do. Here's a practical, scaled-down playbook:

Step 1: Create a simple health dossier. In a document or note, list: key diagnoses and symptoms, current medications and supplements, major lab history (A1c, lipids, thyroid, etc.), and your big health goals (energy, mood, sleep, performance, longevity—whatever matters to you). If you have genetic data from 23andMe or similar, include the raw file or key variants.

Step 2: Feed that context to AI upfront. Start a conversation with something like: "Here's my health summary. Please remember this as my context. I want help building a rational supplement and lab strategy over time." If you're using Claude with Projects or memory features, the context can persist across conversations.

Step 3: Use AI for structure, not diagnosis. Ask it to: map your supplements into categories (core vs. optional), draft a lab request sheet with priorities, explain genetic variants with personalized implications—not generic PubMed summaries.

Step 4: Bring outputs to a human clinician. Print or share your lab sheet, protocol summary, and top questions. Use the doctor's feedback to iterate. AI generates the structured plan; the doctor validates and adjusts.

Step 5: Close the loop. Update your AI with lab results, your doctor's commentary, and your subjective outcomes ("this supplement helped; that one made me anxious"). Let the AI refine the protocol based on real data, always under medical oversight.

The goal isn't to replace your doctor. It's to make you a sharper, more organized patient—someone who shows up with questions that actually matter and context that accelerates the conversation.

Closing Thoughts

This whole process feels different from traditional medicine, and I've been trying to articulate why.

It's not that AI knows more than my doctor. (It doesn't.) It's that AI compresses years of "I'll research that someday" into usable, structured plans. It holds context my brain can't hold. It connects dots across genetic reports, supplements, labs, and symptoms in ways that would take me weeks to do manually.

I'm not passively receiving care anymore. I'm co-authoring my health with AI and my doctor. The AI doesn't replace the human—it upgrades me as a patient. It gives my brain an external R&D department.

Does AI get things wrong? Yes. It can hallucinate, oversimplify, or miss nuance that a specialist would catch. Everything needs verification and humility. But used well, it's the most powerful tool I've found for turning the chaos of personal health data into something coherent and actionable.

This year, when my doctor ordered labs, it wasn't just a ritual. It was the next iteration of a health OS that my brain, my clinician, and a chatbot are building together—one firmware patch at a time.