Coach Roster AI: 2 Rules Justin Harris Keeps Before Automation
AI should widen a coach’s roster without narrowing judgment. The useful test is not whether software can write more check-ins; it’s whether the coach still owns the decision-making where adaptation, appetite, and tradeoffs get messy.
Coach Roster AI: 2 Rules Justin Harris Keeps Before Automation
AI should widen a coach’s roster without narrowing judgment. The useful test is not whether software can write more check-ins; it’s whether the coach still owns the decision-making where adaptation, appetite, and tradeoffs get messy.
Justin Harris says retatrutide at 2 mg produced “no appetite whatsoever” in one client, with more fatigue than usual; the mechanism is appetite suppression. That is a practical AI-coaching lesson because the bottleneck in real coaching is not information delivery, it’s judgment under tradeoff. The sharp thesis is this: AI can scale roster touchpoints and standardize routine output, but if it starts making the tradeoffs for you, coach leverage turns into coach drift.
The leverage problem is real, but it is not the same as coaching
Most of the hype around AI fitness coaching treats the job as a messaging problem: answer faster, check in more often, never miss a week, never forget a detail. Those are real advantages, but they are not the job.
The job is to interpret signals in context. The client who gets weaker because fatigue is rising. The client whose appetite is changing. The client whose food intake is “clean” but whose bodyweight is moving anyway. The client whose insulin sensitivity changes from one high day to the next. Those are not template problems. They are decision problems.
That matters because the best coaching still depends on a human making the tradeoff call. In Justin’s own wording, the retatrutide example is not just “appetite down.” It is appetite down plus fatigue, plus uncertainty about whether the drug belongs in a gaining phase or a lean-out phase. He explicitly says the “helpful in gaining” idea is not something he can reconcile yet and wants more data. That is the right instinct. If a tool changes behavior, the first question is not whether it is impressive. The question is what behavior it changes, by how much, and at what cost.
Where AI actually helps: roster scale without losing the signal
For coaches, the highest-value use of AI is not replacing the judgment call. It is clearing the runway so the judgment call is cleaner.
Think about the repetitive load in a busy roster:
- collecting check-ins,
- summarizing last week’s changes,
- flagging outliers,
- reminding clients what the current priorities are,
- keeping the simple stuff consistent.
AI is very good at that layer. If it can save you 15 minutes per athlete per week across a roster, that is not a small efficiency win. It is leverage. It gives you back enough attention to inspect the cases that actually need a human eye.
That is the real promise: more athletes per coach without turning coaching into a pile of identical responses. The platform should make routine work cheaper so judgment becomes more expensive and therefore more protected.
The failure mode: automation that averages away context
The danger is that AI systems are built to generalize, while coaching often depends on exception handling.
Justin’s off-season framing is a good example. He is not trying to “add muscle” in a simple linear sense. He is trying to teach the body to digest and assimilate a large amount of clean food, reach a place where the athlete can eat more without gaining weight, and build the conditions that support later prep and muscle retention. Whether you agree with every part of that model or not, the important thing is that it is a framework for sequencing decisions over time.
A generic AI coach can repeat the words “high protein,” “progressive overload,” and “consistency.” It cannot, on its own, decide whether this athlete should keep pushing intake, pause appetite suppression, reduce a dose, or hold steady while body composition and fatigue are watched. The minute the system starts making those calls automatically, it stops being leverage and starts being a substitute for the coach’s model.
That is where roster scale can become dangerous. The bigger the roster, the stronger the temptation to let the model flatten the nuance:
- one appetite rule for everyone,
- one recovery rule for everyone,
- one check-in script for everyone,
- one recommendation path for everyone.
But the athlete does not live in the average. They live in the exception.
What to automate, and what to guard
The clean split is simple.
Automate anything that is:
- repetitive,
- low consequence,
- easy to verify,
- and mostly about coordination.
Guard anything that is:
- high consequence,
- context dependent,
- tied to adaptation or recovery,
- or likely to shift the actual plan.
That means AI can draft the check-in summary, but the coach should decide what it means. It can remind the athlete of the week’s instructions, but it should not invent the instructions. It can flag that appetite dropped sharply, but it should not decide that lower appetite is “good” or “bad” without the coach’s framework. It can help you compare this week’s bodyweight, performance, and compliance against prior weeks, but it should not choose between a push, a hold, or a pullback unless you have explicitly encoded the rule and still review the output.
The deeper point is that judgment is not just “being smart.” It is the ability to absorb contradictory inputs and still make a usable call. In the Rory Lazowski exchange, Justin says he’s trying retatrutide himself partly to get first-hand experience for clients, while also noting he doesn’t love the idea of forcing appetite lower. That is not indecision. That is judgment under uncertainty.
The coach’s edge will be model-building, not message-writing
As AI gets better at the message layer, the premium on the coach shifts.
The valuable coach will not be the one who can type the fastest response. It will be the one who can build a better internal model of the athlete and then use AI to execute the boring parts around that model.
That means three things:
-
Use AI to compress information, not replace interpretation. Summaries, trend lines, repeated reminders, and administrative work should be machine-assisted wherever possible.
-
Keep the decision rules explicit. If a client’s appetite plunges, fatigue rises, or bodyweight stalls, the coach should know what variables matter and what thresholds trigger a change.
-
Reserve human review for the edge cases. That is where the money is: the weird week, the mixed signal, the athlete whose plan is working except for the part that matters most right now.
That is also how you preserve trust. Clients do not need a chatbot that sounds confident. They need a coach who can tell the difference between a useful trend and a misleading one.
Bottom line
AI fitness coaching is most useful when it expands the number of athletes one coach can serve without collapsing the quality of decisions. That is coach leverage. But if the system starts optimizing for consistency over context, it quietly removes the coach from the hardest part of the job. The winning setup is simple: let AI handle the roster load, and let the coach keep the judgment.
Sources Used
- raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w13-18m/transcripts/rory_lazowski___members-c5balaovjbdoeefqmfuqdhh2tbpmfdu16lnf0tnrtmw.md
- raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w13-18m/clients/rory_lazowski___members-c5balaovjbdoeefqmfuqdhh2tbpmfdu16lnf0tnrtmw.json
- raw/_consumed/2026-05-26/troponiniq_kb.md
- raw/_consumed/2026-06-02/_GRAS/gras_strategy_training.md