The 3-Part Check-In Triage for AI Fitness Coaching

The 3-Part Check-In Triage for AI Fitness Coaching

Decision quality matters more than “better” predictions when the coach has to act on one weekly update.

The Rory Lazowski retatrutide check-in shows a 2 mg first exposure dropping appetite hard within days, with fatigue showing up alongside it; the mechanism is appetite suppression, not magic. That matters because weekly check-ins are not prediction contests—they are triage. In AI fitness coaching, the winning system is the one that makes the right small decision from incomplete data, fast, and without pretending every signal means the same thing. The thesis here is simple and falsifiable: most coaching errors happen not from missing data, but from failing to prioritize the right check-in signal in the right order.

Weekly check-ins are triage, not diagnosis

A good weekly check-in is not a full-body autopsy. It is a sequence of decisions:

1. Is anything dangerous or clearly off?
2. If not, is the current trend consistent with the phase goal?
3. If yes, do we hold, reduce, or push?

That order matters because AI tools are strongest when the input is structured and the decision rule is narrow. They are weakest when they are asked to improvise around vague language like “I feel off” or “the week was weird.” The coach still has to decide what the signal means.

Justin Harris’s actual coaching pattern in the KB is very consistent with this. In the Joe Webb exchange, insulin sensitivity improved enough that the same insulin dose started dipping blood sugar more than expected on a high day, so the response was not to debate abstract physiology—it was to reduce the dose and bring the meals closer together. That is triage: identify the changed variable, make the smallest useful adjustment, and preserve the plan.

The important lesson for AI coaching is that the check-in is only useful if it can answer a narrow question. Not “How is the athlete doing overall?” but “What changed this week that should alter the next 7 days?”

The first layer: safety or obvious mismatch

The first layer of triage is not glamorous. It is whether the current setup is now mismatched to the athlete’s reality.

In the Rory Lazowski conversation, Justin did not romanticize the appetite drop from retatrutide. He noted the appetite suppression, saw fatigue, and treated the situation as a tradeoff against the upcoming direction of calories and prep. That is the right first-pass frame: if a tool changes appetite too aggressively, the immediate question is whether it helps the phase goal or fights it.

You do not need a grand theory for this step. You need a decision rule:

• If a new intervention causes a clear shift in appetite, energy, or food tolerance, it gets reviewed first.
• If it improves adherence in a phase where lower intake is useful, it may be tolerated.
• If it creates friction in a phase where food volume is the job, it becomes a liability.

That is why “check-in triage” beats “AI insight.” The best weekly system flags mismatches early, before the coach starts overfitting to noise.

The second layer: phase fit

Once the obvious mismatch question is answered, the next question is whether the trend fits the phase.

Justin’s off-season framework in the podcast source is blunt: the point is to teach the body to digest and assimilate a massive amount of clean food, because that supports more potential muscle growth, a better metabolism, and better contest prep later. That means the check-in question is not simply “Did weight go up?” It is “Is the athlete becoming better at handling the amount of food the phase requires?”

That framing is useful for AI because it ties the weekly report to the actual job of the phase:

• In a gaining phase, a check-in that shows stable intake tolerance, manageable hunger, and no runaway weight gain may be a hold.
• In a prep phase, a check-in that shows appetite suppression, increasing fatigue, or food aversion may require a change.
• In either case, the correct decision depends on phase context, not on a generic confidence score.

This is where too many AI products overpromise. They want to produce a clever summary of the week. Coaches need a decision that respects the phase. If the phase is gaining and appetite is crushed, that is not “good adherence.” If the phase is prep and appetite is lower, that may be useful—unless fatigue, training output, or food compliance start slipping.

The third layer: one lever at a time

The most reliable check-ins make one lever obvious.

Joe Webb’s insulin note is a clean example. He did not report a catastrophe; he reported a repeatable change after a high day. The response was not to change five things. It was to reduce the insulin dose further next time. Specific signal, specific lever.

That is exactly where AI can help if it is disciplined. A weekly check-in should be able to parse a short list of variables:

• bodyweight trend
• hunger
• fatigue
• digestion / food tolerance
• training performance
• phase-specific markers like high-day response or meal spacing

Then it should rank them by decision importance. The point is not to maximize commentary. The point is to identify which lever is actually moveable this week.

A coach using AI well should ask: if I change this one variable, what is the most likely near-term effect? If that answer is unclear, the system should default to hold and collect another week of data rather than inventing confidence.

What AI should and should not do in a weekly check-in

AI is useful when it reduces the time cost of sorting. It is not useful when it turns every check-in into a long paragraph of uncertain interpretation.

The practical use case is this:

• The athlete submits a structured check-in.
• The AI extracts the deltas: what changed from last week?
• The AI tags the changes by type: appetite, fatigue, weight, digestion, performance, adherence.
• The coach decides whether the change is phase-positive, phase-negative, or neutral.
• The coach makes one adjustment, or holds.

That workflow is boring, and that is the point. Boring systems make better decisions. They reduce the odds that a coach chases one bad day or ignores a repeated pattern.

The failure mode is also easy to name. When AI is used as a commentary engine, it can make a check-in sound more intelligent than it is. When it is used as a triage engine, it can improve decision quality by forcing the coach to answer the right question first.

The falsifiable test

Here is the test I would use for any AI fitness coaching platform that claims to improve check-ins:

• Can it identify the most important weekly change in under a minute?
• Can it separate phase-positive signals from phase-negative ones?
• Can it recommend one lever, not five?
• Does the coach end up making fewer unnecessary changes?

If the answer is yes, it is helping decision quality. If the answer is no, it is producing better prose, not better coaching.

That is the real standard. Not whether the AI sounds smart. Whether the weekly check-in leads to the right next action.

Bottom line

The KB examples point to the same coaching principle from two directions: appetite suppression and fatigue matter when they interfere with phase goals, and insulin sensitivity changes matter when they force an immediate dosage or meal-spacing adjustment. The mechanism underneath both is straightforward: small physiological shifts can change the correct next decision. AI coaching should be built to catch those shifts, rank them correctly, and keep the coach focused on the single lever that matters this week. That is what good check-in triage does.

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/2026-05-31/kahunas-export/2026-05-31-w13-18m/clients/joe_webb___members-rksigkykimaxwmo_t4_e8nwvbtc2j0etleutkyysads.json
• raw/Justin_on_Podcast.txt
• raw/_consumed/2026-05-26/troponiniq_kb.md