What agentic systems and physical training share is this: they are only as effective as their ability to perceive, evaluate, and adapt within constraints.
Rules fail because they outlive the context they were designed for, not because they’re wrong. Training advice is full of rules. Some say 12 reps for 4 sets. Others say 10 reps for 5. Some say increase the weight until you can only manage 6–8. The rules vary, and the only reason they exist is because they all worked for someone, at some point. There are no universal rules to training, only methods that succeeded under specific conditions. Where else does this logic apply?
In artificial intelligence, an agentic system is one that decides how to act based on feedback. Rather than following fixed instructions, it observes outcomes, updates its assumptions, and adjusts its behaviour over time. A defining attribute of agentic AI systems is dynamic routing. Rather than following fixed paths, agents adjust their actions based on feedback from the environment. They perceive, reason, act, and learn, changing routes or tools when conditions shift. What makes them effective is logic governed by adaptability.
Training works the same way. Coaches can provide direction, but nothing replaces understanding how your body responds to stress through experience, trial, and failure. I’ve trained across rep ranges that were once presented as truth. I made gains training 10 reps for 5 sets at a fixed weight, also with progressive overload. Even low-rep efforts produced results when applied well. The worry becomes: “am I using the most optimal and sustainable techniques that work for my physical environment?”
Results accelerate when goals are defined: reduced body fat, improved definition, increased strength, for example. But strength goals and body-impact goals are not the same. You can barbell squat 200kg without significantly growing your quads because barbell squats prioritize force production and movement efficiency, not quad dominance. That’s where hack squats become more effective. Outcomes also depend on alignment between stimulus and biology.
Early in training, behaviour is mostly reactive. If the weight feels heavy, you drop it. If you’re not sore, you go harder. In AI, this is known as a simple reflex agent, acting without memory or context. This works briefly, then fails because there’s no learning, calibration, or awareness of diminishing returns. With experience, patterns start to form. You remember what caused pain or progress last time and adjust. In AI terms, this resembles a model-based agent, one that uses memory to guide decisions. In training, this might sound like: “Last time I deadlifted 300 pounds, my lower back flared, so I’ll use a belt next time”. This improves safety but often leads to local optimization: speed without direction. Eventually, training becomes goal driven. Decisions are filtered through targets: visible definition, lower body fat, heavier lifts. This mirrors goal-based agents, where actions are judged by whether they move the system closer to an objective. It breaks down when goals don’t align with biology or ignore constraints like genetics. A goal such as “lose two inches of belly fat in three weeks” may be unrealistic for someone predisposed to midsection fat storage. Over time, trade-offs become unavoidable. You recognize that more effort isn’t always better. This is an utility-based agent approach, optimizing for return on investment more than intensity. An example could be: “I could train five times a week, but the recovery debt isn’t worth it”.
But all these approaches eventually reach a ceiling.
Over time, reactive adjustments, pattern recognition, and goal-driven decisions begin to plateau. This is where learning agents differ. Rather than optimizing within fixed rules, learning agents update the rules themselves. Training becomes adaptive, self-correcting, and personal. You stop copying routines and start designing them. This requires honesty, patience, and attention, but it’s where training becomes intelligent and reflective.
The body is not a problem to solve, but a system to explore. There is no single path to strength, adaptation, or performance, only routes tested through feedback. Like agentic systems, progress emerges from iteration: acting, observing outcomes, updating assumptions, and choosing again. What matters is not following the correct rule, but developing the ability to learn from the body over time.