How I Used AI to Get Back In Shape and Lose 30 Pounds

When it comes to AI, many people fall into one of two loud camps. One views the technology as an existential threat. The other sees it as magic—the solution to almost any problem imaginable. What you rarely find is an honest examination like this one, exploring what the technology can truly do, where it helps, and where it doesn’t.

I’m neither an AI evangelist nor a blind skeptic. AI tools like ChatGPT, Gemini, and Claude are resources I use often to make routine tasks a touch easier and more fun. So it felt natural to consult an AI platform or two to help me formulate my fitness plan.

I know some folks will object to the way I approached this exercise, bristling at health metrics like BMI and the scope of my interactions with a widely unpredictable technology, but I’m not a health expert and certainly not advocating that you mirror what I did. I’m just curious about AI and eager to explore what it can do.

There’s nothing groundbreaking here. Lots of other people in search of guidance are doing the very same thing, with good results, according to Laura Richardson, PhD, an exercise physiologist at the University of Michigan in Ann Arbor. “Any way we can help support people to move more is good,” she says. “AI has a lot of great suggestions, and depending on your fitness ability and your intuition about what works for you, it can be a really effective starting point.”

But the technology does have limitations, Richardson adds. AI platforms don’t observe you in real time, so they’re not ideal at offering instruction on proper weight-lifting techniques. And people with conditions such as diabetes, orthopedic problems, or heart disease should not rely on the technology without input from a professional. “When you’re juggling medical complexity, that real-time human coach becomes really important,” she says.

To be safe, she advises anyone working out to watch for three red flags: severe shortness of breath, discomfort in the chest, and lightheadedness. If you experience any one of those things, stop exercising and seek medical help.

By mid-October, I was researching home gyms. I work remotely—no commute to eat up my day—and have a spare room that was largely unoccupied, except for my wife’s treadmill. I started poring over a YouTube channel called Garage Gym Reviews, hosted by a guy named Coop, who tests home resistance-training equipment such as barbells, racks, and benches. (CR tests select gym equipment, such as treadmills and resistance bands, and has helpful advice for those setting up a home gym.)

I tossed follow-up questions at Gemini, comparing various options and pushing for insight on what I actually needed vs. what was upsell. I landed on the Titan Fitness T3, a midtier power rack. After adding a barbell, bench, safety arms, and weight set, the total price came to about $1,700.

I acknowledge the privilege in that statement. Not everyone has $1,700 of discretionary spending, plus a spare room and the flexibility to work from home. What I can say is this: Before moving to Arizona, I spent years working out at YMCAs all over New York City and the equipment I used at those locations was essentially identical to what I built at home. For me, a personal setup is a worthwhile convenience: The nearest gym, an LA Fitness, is about a 30-minute drive away, and I figured if I spent real money on equipment, I’d be psychologically obligated to use it. So far, that logic has paid off.

On Nov. 1, exactly one month after my first walk with Brian, I asked Gemini to design a lifting routine for me. I provided it with a few key details on my goals, including the fact that I was consuming around 160 grams of protein daily and wanted to train four days a week. It built me an upper- and lower-body split—Upper A, Lower A, Upper B, and Lower B—on a two-days-on, one-day-off cycle, designed around my equipment and needs. Bench press, squats, overhead press, deadlifts, barbell rows. Nothing too clever; just basic compound movements performed consistently, with slightly more weight or reps added each week.

I tracked every lift in a log (using the iPhone app RepCount). I weighed myself daily and recorded the numbers without getting too hung up on them. I tried different foods to see how they affected my hunger and energy levels. I treated my body as the product under review, and the data as the only input that mattered.

For good measure, I added a second AI tool—Anthropic’s Claude—to the rotation to double-check what Gemini was telling me. The two digital instructors developed distinct roles, almost naturally. Gemini was the architect: mapping strategy, designing the program, and celebrating milestones with what approached genuine enthusiasm. Claude was the quiet analyst: confirming the math, stress-testing my assumptions, and evaluating my progress.

Every few weeks, I’d upload photos of myself to Claude—same lighting, same front-and-side poses—and ask for a read on things: Was I losing fat or muscle? Where was the change most visible?

Let me add a measure of caution here: Digital tools like this have been known to promote unrealistic physical ideals and respond insensitively to people with body image concerns. What’s more, they’re not bound by the same strict privacy laws as your doctor’s office. For the sake of this exercise, though, I opted to give this a try, uploading standardized photos and asking specific clinical questions.

I’ll let you judge the results.

In one early session, I pulled up a photo of myself from 2020, when I was lean from cycling. Claude’s assessment was measured: The 2020 version was lean, yes, but without proper protein consumption and progressive overload—gradually pushing my body to do more work than it’s used to—there wasn’t much muscle there.

“I remember taking that photo and thinking I looked pretty good,” I said. “I guess I was lean but not especially big.”

Claude agreed and explained how this time would be different.

I also laid out my individual goals, which were, in my mind, relatively modest: feel healthy, fit into some old clothes, and look reasonably like someone who exercises. No six-pack ambitions, no fitness influencer delusions—just a 40-year-old trying not to feel like a stranger in his own body.

Claude confirmed the math, walked me through a realistic timeframe, and helped me understand the long, steady weight reduction my strategy would require.

“Assuming I stay consistent,” I asked at one point, “is this mostly just a matter of time?” I used to live in New York City and watch skyscrapers go up over a two-year span. The blueprint changed very little; the key variable was time.

Claude agreed once again. That framing stuck with me every time the bathroom scale slowed to a virtual halt.

Back in the Gemini thread, the coaching energy was high. When I asked about the pull-up bar on my rack, Gemini added pull-ups to the routine—and addressed a looming problem: In my current state, I could successfully complete approximately three-fourths of a pull-up. I acknowledged this with a sheepish “lol.”

The response: “That ‘lol’ is mandatory when you feel like you’re pulling for your life and your chin just won’t clear the bar.”

Gemini laid out a progression protocol, using my bench as an assist—jump to the top position, lower yourself as slowly as possible in a three- to five-second descent, building strength from the top down. “Week 5 or 6,” it predicted, “you’ll go for your max-effort pull, and your chin will just ... go over the bar. You will have just done your first clean pull-up.”

About six weeks later, that’s exactly what happened.

Throughout November and December, I was checking in with both tools after workouts, after weigh-ins, and whenever something felt off. When I reported that my scale weight was barely moving, but my lifts were going up and my waist was visibly shrinking, Gemini explained the underlying math: If you burn 10 pounds of fat while simultaneously adding five pounds of muscle, the scale only moves five pounds—but your body has fundamentally changed.

“You are trading fluff for armor. This is the Holy Grail scenario.”

The “fluff for armor” line sounded a little over the top, but that’s AI for you: eager to please.

Presented with the same numbers, Claude quietly confirmed that thinking: The tape measure and workout log were the true arbiters of success. The bathroom scale was the least reliable narrator in the room.

In mid-December, my waist dropped 1.25 inches in just 10 days. Gemini responded with the energy of a coach pounding a whiteboard: “1.25 inches in 10 days? That is not just progress—that is a collapse of the enemy lines.” It explained the physiology: deep core muscles, finally activated by weeks of deadlifts and planks, tightening like an internal corset. 

Is that really what happened? I can’t say for sure. All I know is that my plan was working.

I tried not to get too hung up on the numbers.

“You are going to break the 40-inch barrier before Christmas,” Gemini said. “That is going to be one hell of a gift to yourself.” I appreciated the sentiment, but reading that line again now does feel needlessly saccharine.

By January, the changes were no longer subtle. My arms had definition. Pants I’d written off were fitting again—and then getting loose on me. I shared a new batch of photos with Claude alongside that 2020 cycling photo. The contrast was obvious. I’d been lean in 2020—but only because I’d cycled the fat off without adding anything to replace it. Things were different this time around.

I found myself thinking this whole experience might make for a good Consumer Reports story. “I’m just a nerd who likes old video games,” I told Claude. “If I can get in shape at 40, then anyone can.” What I’d come to understand—through five months of data logging, AI conversations, and shifts in body mass—is that this story isn’t about artificial intelligence at all. It’s about finally understanding how things work.

In February, I got hit with an unexpected twist. I contracted COVID-19. And a week after recovering, I was goofing around with Brian and injured the big toe on my right foot, suffering a contusion severe enough to take walking, let alone training, off the table. For three weeks, I didn’t lift, didn’t venture outside, didn’t do much of anything. It was not fun, let me tell you! I expected to lose significant ground.

I didn’t. When I circled back to Claude, it walked me through why: The system was sound enough to survive the interruption. The muscle wasn’t gone. The knowledge wasn’t gone. Starting back at roughly half my normal weightlifting limits, just getting into rhythm once again—it felt much like it had before.

By early June, I had dipped to 160 pounds, a 37.5-inch waist, and roughly 19 percent body fat. I would probably have been a smidge lighter if I hadn’t hurt my toe, but it’s not a big deal. My bench and deadlift hit 185 pounds. Those may not be impressive numbers if your name is Brock Lesnar, but I write about laptops for a living. I’ll take it.

Before wrapping up, let me share a few last tips from Richardson.

For strength training, she recommends what exercise scientists call “reps in reserve”—the idea that you should always leave a couple of reps in the tank rather than pushing to extremes, particularly when you’re lifting without a spotter. “It’s like gas in a gas tank,” she says. “How much do you still have left? That’s your gauge.”

She also offers practical advice on how to effectively use AI from the start. Rather than simply asking ChatGPT or Gemini to write you a program, she prefers a more conversational approach: Share who you are and what your goals are, and then explicitly ask the tool to interview you before it generates anything. “Say, ‘Before you write the program, ask me 5 to 10 questions that help you better understand who I am and what my goals are.’” The AI will generate those questions, you respond, and the result is a more personalized program. “That’s what people neglect to do sometimes,” she says. “You just go in and ask the question instead of trying to really prime the tool.”

And, finally, she endorses the practice of logging your behavior—food, exercise, sleep—and periodically uploading that data to AI for review. “You can go back weeks later and say, ‘Here’s what I’ve been doing. What suggestions do you have?’” As the journey progresses, she adds, the relationship with the AI naturally evolves: At the beginning, it’s about building a program; later, it can help you troubleshoot plateaus, stay motivated, or find ways to fit fitness into a changing schedule.

I’d like to weigh around 150 pounds by Labor Day or shortly thereafter, then shift to a slow, deliberate bulk—eating around 200 calories a day above maintenance to add muscle through 2027. It’s a long-range strategy, and I intend to follow it for the rest of my life, now that I finally know what I’m doing and why.

Also, my clothes fit again. That’s nice, too.