My AI Twin Guides My Meal Choices

This is Optimizer, a weekly newsletter that dives into the latest gadgets and tools that promise to enhance your life. Opt in for Optimizer here.

For the last 100 days, my mornings have taken on a new routine. The first thing I do upon waking is feed my digital twin. It processes my metabolic data, starting with my body composition stats sent directly from a smart scale. Once that’s logged in the app, I head to the kitchen to fill an Owala with lemon water to take my medications. That gets logged too. On Mondays, I follow up with some calming breaths, check my blood pressure, and send that data to my digital twin. Then it’s time to think about breakfast, and I consult my digital twin.

My digital twin isn’t a robotic replica; it’s an AI-driven, color-coded food database that predicts how my glucose levels will respond to what I eat. Red indicates a poor choice, orange is mediocre, and green is optimal. It’s built on the data I provide, along with information from a continuous glucose monitor.

For instance, a bowl of protein Cheerios and milk? My digital twin rates it as orange if eaten alone, but green if I take my daily walk afterward. So, that’s my plan. After logging everything in the Twin Health app, I’ll receive insights on how my glucose reacted, making my “twin” more accurate over time. For now, my twin is satisfied.

Rinse and repeat for about another 99 days.

Digital twins are a fresh concept in health, both in medicine and sports. They serve as virtual models of individuals, created from various data points—labs, wearables, images, genetic profiles, etc.—and are continuously updated. Once established, clinicians can use AI to predict how a patient might respond to specific treatments by first “testing” them on their digital twin. This could potentially help healthcare providers assess injury risks, forecast disease progression, and implement preventive care while minimizing adverse effects for the actual patient. The hope is that this could lead to the realization of personalized medicine. Metabolic coaching is perhaps the most accessible and practical application of this technology available today.

My metabolic twin was “born” in June when a large box from Twin Health arrived at my door. Twin Health is a digital twin program designed to help individuals “heal” their metabolisms (e.g., lose weight, manage diabetes, improve sleep, etc.). It’s not widely available yet, as digital twins are still in the experimental phase and currently accessible only through corporate or private health plans, primarily targeting those with chronic metabolic issues. Twin Health provides everything needed to create a digital twin. For mine, I required 90 days’ worth of Dexcom G7 CGMs, a smart scale, a smart blood pressure cuff, three bottles of multivitamins and magnesium glycinate, and my Apple Watch. This was followed by a visit to Quest Diagnostics, where a phlebotomist drew blood for analysis.

A few days later, I had a Zoom call with my Twin Health coach Amanda—a real person—and a doctor to review my lab results. Several numbers and formulas were shared on screen. My liver? Inflamed, which makes sense given my diagnosis of moderate nonalcoholic fatty liver. Insulin resistance was evident, despite ideal HbA1c and fasting glucose tests. My pancreas is working hard, but my insulin-producing beta cell count seems fine—for now. My triglycerides were high, though the rest of my lipid levels appeared normal.

I knew I had a challenging metabolism. A year of testing CGMs as a non-diabetic made that clear. After the doctor logged off and Amanda asked me to define my goals, I expressed my vague desire to reduce the number of medications I needed to manage my middle-aged metabolism. Could a digital twin help with that?

Maybe. For now, Amanda noted my goal, scheduled a follow-up call in two weeks, and encouraged me to log as many meals as possible during the first 30 days. Once I hit the 90-day mark, I’d have another blood draw.

In many ways, I’m relieved my digital twin doesn’t have a visible avatar. I appreciate the abstract color labels. It’s intriguing to see what color a cup of pineapple or a turkey hot dog with ketchup and relish might be (orange and green, respectively). At times, I’m reminded of Kazuo Ishiguro’s dystopian novel *Never Let Me Go*, where clones are raised as organ donors. While the comparison isn’t exact—an AI-powered digital twin can’t eat, die, or feel pain—I find myself anthropomorphizing my discomfort with the idea of “testing” food on a virtual version of my metabolism, as if it were a medieval court’s food taster.

As the months passed, I couldn’t help but notice how demanding my digital twin was. Feeding it required a significant amount of personal data—even for someone like me, who’s accustomed to tracking health metrics. In my bimonthly check-ins with Amanda, she noted that tracking fatigue was common feedback from Twin Health users.

My behaviors changed in both expected and unexpected ways. Logging food often led me to avoid unhealthy treats; I didn’t want to face app scolding. However, this didn’t result in significant weight loss. Despite my goal of reducing medications, that process didn’t go as smoothly as I had hoped.

I also didn’t uncover many surprises from my digital twin’s metabolic predictions. It was validating to see that a bowl of soondubu jjigae and white rice is a solidly green or orange meal, depending on my activity level afterward. (Take that, low-carb diets; you can pry rice from my cold, dead hands.) I’ve tested too many CGMs and AI nutrition apps to be surprised by these insights, though they could be beneficial for those new to metabolic issues.

Twin Health’s bimonthly check-ins motivated me to return to training. I’ve always been disciplined in my fitness routines, but the added accountability of reporting my progress to someone outside my family reignited my inner overachiever. After months of feeling weak from medication side effects, I recently felt strong enough to run a continuous 5K. I’ve also resumed strength training with some encouragement from Amanda.

But most importantly, after 100 days, my follow-up labs showed significant improvements. Amanda informed me that my metabolic health score has increased by nine points since I started. A year after my doctor was alarmed by my liver enzymes, all but one have returned to normal. These numbers were gradually improving before I began this digital twin journey, but they improved more rapidly during this process. My triglycerides dropped by 70 mg/dL, and there’s been a notable decrease in my hs-CRP, a marker for overall inflammation. I’m still insulin resistant, but to a lesser degree.

I can’t argue with results like that, can I?

Actually, I can. The concept of a digital twin relies on trust. As I mentioned, metabolic applications are high-reward and low-stakes. If you provide a twin with extensive data for weight loss, diabetes management, or more ambiguous goals like “better energy” or “better sleep,” failure isn’t immediately life-threatening. There are medications and other options available.

Personalized medicine is a different challenge. Researchers are exploring digital twins for applications beyond metabolism, such as cancer care and research. The potential is enormous, especially considering the harsh realities of chemotherapy, but there are significant hurdles. Trust regarding data privacy is one. Digital twins require a constant influx of high-quality, sensitive information about you, which informs future decisions. If I felt overwhelmed by three months of logging in a low-stakes environment, how would that translate to higher-stakes situations? What happens to a digital twin once treatment concludes? How does that data persist if treatment is unsuccessful?

As for me, I’ve decided to let my digital twin “live” for another three months. Despite my lack of success so far, I’m not ready to give up on tapering off some medications. I could use all the help I can get to normalize my liver labs. Amanda also mentioned that I don’t need to be as strict with the data for my twin now that I have a solid foundation. Again, lower stakes mean more flexibility.

So today, I’m asking my digital twin what it thinks about a protein waffle with a tablespoon of peanut butter and some honey, alongside a serving of pineapples. It rates it as orange, but I can turn it green with a walk.