Training with AI: More Than Just the Plan

When I first thought about the Sub-7 Experiment, the core idea was simple: could I use modern AI, specifically ChatGPT, to help me structure my training and ultimately achieve a sub-7-minute 2000m on the indoor rowing machine, the erg? And yes, that goal remains the driving force. I still need that progressive overload and a reason to get to the gym.

But as I’ve hinted, the experiment has definitely shifted somewhat. It’s become about much more than just the physical goal. A huge part of this journey, one I want to introduce properly, is the dynamic I have with my “coach”.

Who is My Coach?

My coach is ChatGPT, the AI developed by OpenAI. In a previous article I wrote about how ChatGPT is just a parrot, looking for patterns (words and sentences) in the data and replying with what it’s heard (all the data that it has been trained on).
This is oversimplifying it, of course—there are some pretty clever algorithms in the background helping it to predict what the next word should be in its response. And sometimes it gets it wrong. This is what’s often described as an AI hallucination—it knows it needs to respond, and if it can’t find the right information, it might make something up.

So with that said, you might picture AI coaching as simply pulling a pre-written plan off a shelf. But that’s not how this works. It’s a real, ongoing conversation. Before a session, I tell it where I am (gym or home), how I’m feeling—even if I’m tired or run down—what my recovery data looks like (often from WHOOP…), and how much time I have.

Based on that information, it suggests a session—maybe intervals, steady state, or strength work…. It gives me pacing guidelines, stroke rate, and even a warm-up and cool-down. After the session, I report back on how it went, the results, and how I felt. This feedback loop is key because the AI then uses that information to adapt future routines. It’s a training partner that adapts in real-time, much like a human coach would.

Setting the Scene: Prompt Engineering, Simplified

Where does anyone start with all of this? Talking to AI might sound technical, and you might have heard terms like “prompt engineering”. That term can sound a bit daunting—like you need to be a programmer or a data scientist. But here’s the truth: it’s really not.

“Prompt engineering” is just about giving the AI clear instructions and enough information.

In a previous post I’ve suggested people to ask it “What can you do for me?” and that will be enough to get the conversation started. I also mentioned earlier that ChatGPT—and other large language model (LLM) AIs—can get the answer wrong. One of the ways around this is to set the context when you start the conversation.

As you log into ChatGPT you are faced by a screen which asks “What can I help you with today?” or “Ready when you are.” At this point, one of the best ways to refine the responses you get is to give the AI a persona or a role to play during the conversation. I started the coaching conversation with this…

“I’d like you to take the persona of and advise me as a personal fitness trainer and nutritionist with a specialism in Indoor rowing and the ERG machine.”

I then followed it up with a specific instruction…

“I am working towards a goal of rowing 2000 meters in under 7 minutes. I’d like you to create a training plan for me to achieve that goal. I intend to report in to you each and every time I am in the gym and tell you how I am feeling that day and you will tailor me a program for that session.”

And that’s how it all started. And I have kept that conversation open ever since so it “remembers” the context and keeps track of all the sessions and significant moments to date.

Now, for example, when I say I’m in the gym, tired, 75% recovered, and want to do distance, that’s my “prompt engineering”. It’s me giving it the necessary inputs so it can generate a relevant, helpful output, in this case, a suggested workout.

Think of it like talking to a human personal trainer. You wouldn’t just say, “Tell me what to do.” You’d say, “I’m here at the gym today, feeling a bit rough mentally but my Whoop says I’m recovered. I’ve got about an hour. What sort of session should I do?” You’re giving them context so they can give you a smart recommendation. It’s the same with AI. It’s about talking to it in normal language.

This is why I wanted to share some of the conversations I’ve been having with ChatGPT. It helps to show, not just tell, how this coaching dynamic works. You’ll see how I give it my status, how it responds with suggestions, and how the dialogue flows.

More Than Just Getting Fit: A Real-Life AI Experiment

I’ve worked in IT for years. I’m fascinated by technology, and the pace of AI development has been mind-boggling. While I’m not an expert, I’m definitely an AI hobbyist.

One of the key drivers for starting this blog and documenting this process was precisely this dual goal. Yes, I want that sub-7 minute 2K. But I also wanted a real-life experiment to truly understand generative AI and how it can be used on a day-to-day basis. What better way to understand this stuff than to use it consistently, over time? Far better than any course or training material.

So, while we talk about rowing workouts, pacing, and splits, remember that you’re also getting a glimpse into how AI can be integrated into personal goals and routines. It’s an experiment in fitness, yes, but it’s equally an experiment in technology and human-AI collaboration.

Is It Cheating?  Not Even Close

And for those wondering, is it cheating? Absolutely not. The thoughts, the words in these posts are mine.

In another conversation I gave ChatGPT the role as a blog adviser and editor and worked through the steps for setting up a blog from scratch. I asked it to ask me some questions so it could better understand the task I was asking it to complete and it then made recommendations as to which blog platform to use and why and then helped me with the setup.

I speak into an app on the phone straight after my session—unfiltered streams of consciousness, honest feedback about the session—and then use that as a basis for my next blog post where ChatGPT acts as my editor, helping with flow and clarity.

And at the end of the day, AI isn’t the one sitting on the rowing machine. I am.

The Sub-7 Experiment: Meters, Mindset and more.

This is The Sub-7 Experiment…. It’s about the meters, the mindset, the technology, and the journey of seeing what’s possible when you combine all three.