Swimming with an AI Coach:
Total Immersion Meets Deep Research
This summer, I’m embarking on a long-distance swimming journey – but I’m not doing it alone. My training partner isn’t a human coach or a fancy swim app; it’s an AI. Specifically, I’ve recruited OpenAI’s new o3-pro model as my personal swim coach, supercharged with its agentic Deep Research to make it even smarter. This post kicks off a series about how I’m building an AI-driven training partnership to get back in shape in the pool. It's about how genuis-level AI can help one out-of-practice swimmer (👉 me) train smarter, adapt on the fly, and maybe even have some fun along the way.
Back to the Pool (After 5 Years!)
It's been nearly 5 years since I last really swam laps. I’ve splashed around with my kids here and there, but proper training? Zero. This morning, I took in the glorious California summer sun and went to our local outdoor pool. 1,000 yards (of 50 meter laps) later, I was exhausted but reminded of how much I can actually like swimming (sunlight and an open-air faraday cage? Perfection.) My arms felt like noodles, my lungs were on fire, and I had the humbled grin of someone who just remembered that swimming shape is ... different.
My Apple Watch doesn’t lie: 1,050 yards of swimming (including a modest 50-yard butterfly attempt) took me just under an hour – with an average pace per 100 yards that a sea turtle would laugh at (or not, they're such beautiful and kind creatures). Seeing those numbers after my first swim back was equal parts humbling and hilarious. Maybe an AI-savant can help?
If you want a laugh, check out my splits: my first 100 yards (with generous rest included) clocked in at a glacial 7 minutes 37 seconds. 🤦♂️ By the time I got into a rhythm, I was still slogging along, and my heart rate was spiking into the 160s. The graph looks like a seismograph during an earthquake – and I'm not sharing my VO2max to avoid embarrassment. I have nowhere to go but up from here, and that’s where my new AI coach comes in.
Why an AI Coach (Instead of a Human)?
After replacing Otter.ai with 4 prompts, I wondered how GPT might help me here. I didn’t just want a generic plan; I wanted a coach that knows me – my goals, my schedule, the fact that I’m juggling work and family, and that I’m motivated by fun (and cool tech) as much as by getting faster.
So, I gave the AI all the personal context it could need. I told it my current fitness level (screenshots from today's swim), my long-term goal (build endurance for long-distance swims by end of summer, and not drown when I race my kids in the pool), and my constraints (e.g. I can realistically swim 3 times a week, I have ~45 minutes per session, and learning to tread water like a water polo player sounds like it'd be handy for pool time kid play). I basically handed over a dossier on “Sutha, the Wannabe Swimmer.”
Then, I handed the AI some expert knowledge to work with: I uploaded Total Immersion by Terry Laughlin – the classic book on efficient swim technique. Yes, I bought the book (probably more than once), so I'm happy to hand the AI a PDF of it. Why that book in particular? Because I knew I didn’t just need endurance, I needed to relearn how to swim well. Terry Laughlin’s philosophy is that well-executed technique beats brute fitness any day. In other words, flailing through endless laps won’t help if you’re swimming like a brick. His Total Immersion method preaches balance, streamlining, and efficiency: swim smarter, not harder. Deep Research noticed when scouring the web that there lots of folks who preach "just keep swimming until you finally get a continuous mile." Maybe that works for some folks, but isn't my style, so giving it my "source of truth" text was key.
Finally, I turned on Deep Research mode for the AI. What’s Deep Research? Think of it as giving my AI coach a superpower: it can actively read and synthesize info from the internet and my files instead of just regurgitating whatever it was trained on. Equally important, it's an "agentic" mode: it means the AI can scour sources, delegate subtasks to itself (spinning up as many instances of its genius-self as it needs), and plan with real data and citations, almost like a research assistant. This isn’t your vanilla ChatGPT spouting generic tips – it’s an AI that can double-check itself, pull in new ideas, and think more deeply about my training plan. There's no question it's smarter and more thoughtful than the median human coach.
Building the Plan: Total Immersion x AI Adaptation
Within hours of feeding it all this info, my AI coach – I should really give it a nickname – came back with a draft training plan. It felt spot on. The plan wasn’t about throwing me into some Navy SEAL bootcamp. Instead, it started gently: lots of drill work, technique focus, and low-key endurance building. Essentially, it told me to slow down to eventually speed up (music to a nerd’s ears). Here are a few things that immediately made me smile in the plan:
Balance and “Pressing the Buoy”: The AI prioritized drills from Total Immersion that improve my balance in the water. One of the key tips is “press your buoy” – which basically means pressing your chest (your buoy) down a bit to get your hips up. This makes you more level in the water and reduces drag (you almost feel like you’re swimming downhill when you get it right) . My plan literally has Day 1 focusing on balance drills, finding that sweet spot where I can float effortlessly without kicking like a maniac.
Streamline and Stretch Out: Another focus is on swimming taller and slippery. The AI coach reminded me (with quotes from Terry Laughlin’s book and all!) that a longer body line cuts through water better. So, there are drills for extending my glide, and keeping a front-quadrant stroke – fancy talk for not starting the next stroke until the lead arm has had a moment to glide. Quality over quantity is the name of the game; the plan has me doing slow, deliberate strokes to retrain my muscle memory. As Laughlin noted, efficient technique naturally leads to endurance and speed gains over time .
Kicking Back (Literally): My AI even worked in some kick drills and suggested fins for certain sessions. I have a love-hate relationship with kick sets, but it knows (from my input) that my leg endurance is a weak point. Still, the emphasis isn’t on generating huge thrust with my kick; it’s about a relaxed, minimal kick that keeps me balanced. In Total Immersion style, the kick is just to stabilize, not to propel me like a motorboat. (Praise be; because if it asked for 1,000-yard kick sets, I’d promptly quit 😅.)
Adaptive and Accountable: How the Series Will Work
This blog series will follow my AI-human training experiment as it unfolds. Here’s how it’s going to go down: after every week (or key workout), I’ll be logging my data and feedback to the AI. That means if I crush a session and need more challenge, the model will crank up the difficulty a notch. If I’m struggling (looking at you, 7-minute 100s 🥵), it will adjust and dial things back or tweak the technique focus. The plan is fluid and adaptive, just like a good human coach would be – only my coach lives on a server and dreams in high-dimensional embeddings (man, robot sheep sound so much easier to understand).
What’s fascinating is how interactive this process is. Using the o3-pro model with Deep Research, the AI isn’t just passively taking notes; it’s actively analyzing what I log. For example, if I report that my breathing is what’s limiting me (pretty likely, since right now I feel out of breath quickly), it might dig into its sources and come back with a couple of new breathing drills or tips. If my shoulder starts aching, it could pull up some injury prevention stretches or modify my workouts to avoid aggravation. It’s like having a coach who has instant access to all the training literature and can personalize it on the fly – a coach with an eidetic memory and a knack for data analysis.
Another big plus: accountability. There’s something surprisingly motivating about telling an AI how your workout went. It’s objective – it doesn’t let me off the hook easy, but it also doesn’t judge. If I skip a workout, trust me, I’ll hear about it from my digital coach (in the most politely analytical way possible, unless - I guess - I tell it how to abuse me in just the right way). And because it’s tracking my progress closely, it can show me cool charts or summaries of how far I’ve come.
The Road Ahead
So here I am, at the start of Summer 2025, partnered with an AI to become a better swimmer. It’s equal parts exciting and experimental. Will this AI-driven approach actually give me an edge compared to just following a plan from a book or a human coach? Will I stick with it when the going gets tough?
In future posts (presuming I don't drown?), I’ll share how the training plan evolves, the workouts I’m doing, and the breakthroughs (or near death experiences) along the way. I’ll let you know what it’s like handing over the analytics and planning to an AI – the good, the bad, and the funny. And of course, I’ll report on whether my pace improves from “sea turtle” to at least “dolphin-ish.” 🐬
If you’re tech-savvy and fitness-curious, or just want to see a nerd turn himself into a swimmer with the help of a super-smart AI, I invite you to follow along. This isn’t just about me trying to swim faster; it’s about exploring a new frontier of training where human experience and AI intelligence meet in the lane lines. I’ll be back with updates – and maybe even some data charts and more war stories from the deep end. Stay tuned, and see you at the next turn!
Thanks for reading! Dive into the comments if you have questions or if you’ve ever tried something wild like an AI coach. And if you have a better name for my o3-pro swim buddy, I’m all ears. 😉