Weekly Check-In Triage: One Blood Glucose Signal, One Appetite Signal, One Decision

Weekly Check-In Triage: One Blood Glucose Signal, One Appetite Signal, One Decision

AI coaching is only useful when it improves the next weekly call; the mechanism is change detection, not prediction.

The strongest decision rule in the Kahunas coaching corpus is simple: when a high-day insulin dose starts dipping blood sugar earlier than usual, Justin Harris cuts the dose immediately and brings meals closer together. That is a weekly check-in triage mechanism, not a vibe check — observe the change, identify the constraint, act on the constraint. The falsifiable thesis is this: AI coaching wins when it flags a meaningful state change fast enough to change the week’s plan, and loses when it turns check-ins into generic reassurance.

The weekly check-in is a triage tool, not a progress report

Most coaching failures come from reading the wrong signal too slowly. A weekly check-in is not where you write a biography of the athlete’s week. It is where you answer three questions:

1. Did something materially change?
2. What is the narrowest explanation for that change?
3. What action keeps the plan moving with the least collateral damage?

That is exactly how Justin handles real check-ins in the corpus. When Joe Webb reported that his usual high-day insulin dose now caused a noticeable dip in blood sugar, forcing meal 2 to move up 30 minutes, the response was not to “wait and see” for another week. The immediate action was to reduce the dose further on the next high day. The logic is obvious: the signal was specific, the effect showed up under a known condition, and the correction was local. No drama, no overfitting, no pretending the dose still fits because last week it did.

That is what good triage looks like in coaching. You are not deciding the athlete’s entire future from one check-in. You are deciding whether the current script still matches the current physiology and schedule.

Appetite suppression is only useful if the macro plan can absorb it

The same triage rule shows up in Justin’s retatrutide comments to Rory Lazowski. Rory described a 2 mg dose on Friday, then “no appetite whatsoever” through the weekend, plus more fatigue than normal. Justin’s response was not to celebrate appetite suppression as a universal win. He immediately framed the tradeoff: he did not love the idea of forcing appetite lower, but he tried it for firsthand client experience; then, when body composition allowed it, he recommended leaning out while appetite suppression was making that easier.

That sequence matters. The appetite signal is only helpful if it is matched to the phase of training. If calories need to rise, a stronger appetite blocker can become a liability. If calories are still coming down anyway, the same effect can be useful. The coaching decision is not “is the compound good or bad?” The decision is “does this week’s appetite state help or hurt the next required move?”

That is the difference between decision quality and mechanistic fascination. A lot of tech stacks, AI included, are great at generating commentary about the signal and mediocre at deciding what to do with it. The useful triage system asks whether the athlete is in a gaining phase, a mild cut, or a prep ramp, then treats appetite suppression as conditional rather than heroic.

Check-ins should prioritize constraint changes, not total data volume

Justin’s answers are effective because they reduce a noisy week to the constraint that actually changed. In Joe’s insulin example, the key constraint was not bodyweight, mood, or food choice. It was insulin sensitivity on high day. In Rory’s retatrutide example, the key constraint was not whether appetite reduction is interesting in the abstract. It was whether the athlete’s current phase could tolerate lower intake.

That same pattern appears in Justin’s broader nutrition coaching: the details matter, but only after the macros are controlled. In the nutrition knowledge base, he says he would be surprised if fruit choice alone produced a noticeable difference over a year if macros are otherwise nailed; the big results come from getting the macro structure right, while the last few percent lives in the details. That maps cleanly onto check-in triage. First identify whether the macro structure is intact. Then decide whether the smaller lever deserves attention.

AI coaching should copy that hierarchy. If a weekly check-in floods the coach with everything — sleep quality, stress, pump, digestion, scale weight, appetite, step count, “felt flat,” “felt good” — it should not blur the actual decision. The model’s job is to surface what changed enough to alter the plan. If it cannot rank those signals, it is decoration.

The best weekly checks produce local moves, not global resets

The pattern across the KB is local adjustment. Reduce insulin by 1 IU. Pause or reduce the retatrutide dose if food intake needs to rise. Keep fruit where it makes sense and stop obsessing over microscopic carb-source differences when macros are already stable.

That matters because global resets are expensive. They create new uncertainty, new compliance problems, and new opportunities for the athlete to blame the plan instead of the week’s actual condition. Local moves preserve continuity. They keep the training cycle intelligible.

This is where AI can help if it is disciplined. A good weekly check-in triage layer could do three things well:

• detect a deviation from the athlete’s prior baseline,
• classify the likely constraint domain, and
• recommend the smallest change that preserves the plan’s intent.

That sounds basic because it is basic. But basic is where the value is. The athlete does not need the system to invent a strategy. The athlete needs the system to notice when the current strategy stopped fitting.

Decision quality beats narrative quality

A bad check-in often feels productive because it contains lots of explanations. A good check-in often looks boring because it just tells you what changed and what to do next. The Rory and Joe examples are both boring in the best way: one variable moved, the coach named the mechanism, and the next action followed directly.

That is the standard AI fitness coaching should meet. If the software cannot distinguish between “appetite is low but calories still need to go up” and “appetite is low while leaning out is the immediate task,” it is not improving decision quality. If it cannot notice that a previously tolerated insulin dose now pushes glucose down too far, it is not triaging. It is logging.

The sharp claim is this: weekly check-ins are only valuable when they change the next action, and the best AI systems will be the ones that make those changes smaller, faster, and more phase-appropriate. Anything else is just formatted noise.

What coaches should look for

If you are using AI in a coaching workflow, judge it by the quality of its triage, not by how fluent it sounds.

Look for whether it can:

• identify the one signal that changed,
• explain the mechanism in one short phrase,
• say what part of the plan is now out of date,
• and recommend a local correction instead of a total rewrite.

That is enough. You do not need a model that sounds wise. You need one that spots when the athlete’s week has crossed a decision threshold.

When weekly check-ins do that, they become a real coaching tool. When they do not, they are just status updates with better grammar.

Sources Used

• raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w13-18m/clients/joe_webb___members-rksigkykimaxwmo_t4_e8nwvbtc2j0etleutkyysads.json
• raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w13-18m/transcripts/rory_lazowski___members-c5balaovjbdoeefqmfuqdhh2tbpmfdu16lnf0tnrtmw.md
• modules/03-knowledge/kahunas-coaching-deep-nutrition.md
• modules/08-voice/kahunas-coaching-deep-voice.md