Check-In Triage and the 1-Week Insulin Read: A Decision-Quality Problem

Check-In Triage and the 1-Week Insulin Read: A Decision-Quality Problem

Why weekly AI coaching lives or dies on whether it notices the right signal, not whether it can write a prettier summary.

Justin Harris’ check-in logic is brutally practical: when a client’s high-day insulin sensitivity improved enough that the same dose caused a noticeable blood sugar dip, the answer was not to admire the trend — it was to reduce the dose and bring the meals closer together immediately. The mechanism is simple: feedback sensitivity. In weekly coaching, the best AI is not the one that speaks most confidently; it is the one that triages the check-in correctly and acts on the first reliable signal. That makes a falsifiable thesis: in physique coaching, decision quality on the weekly check-in matters more than raw messaging volume, because a good triage system catches the small change before it becomes a bad week.

Why check-ins are not status reports

A lot of coaching tech is built like a dashboard: collect photos, weight, sleep, steps, maybe a mood score, then spit out a neat summary. But the actual job of a weekly check-in is narrower and harder. It is a triage event.

The question is not “Did something happen?” Something always happened. The question is:

1. Is this a normal fluctuation or a real directional change?
2. Does it need an immediate adjustment, a delayed adjustment, or no adjustment?
3. Which lever is the right lever?

Justin’s responses in the KB are consistent on this point. He does not treat every report as equally actionable. He reads the situation, identifies the mechanism, and makes a decision call. In one example, a client says appetite is crushed after taking 2 mg of retatrutide and that fatigue is higher than normal. Justin’s response is not “great, lower appetite means better compliance.” It is more cautious than that: he says he is trying it himself for firsthand experience, but he does not love the idea of forcing appetite lower, and he holds back on celebrating the effect until he has more data. That is not indecision. It is disciplined triage.

The same pattern shows up in nutrition execution. On high days, when insulin sensitivity improved, the client noticed a similar insulin dose now dipped blood sugar enough to require the next meal 30 minutes earlier. Justin’s response was to reduce the dose further on the next high day. No drama, no overcorrection, no worship of the original plan. The signal was clear, and the plan changed.

The real job of AI coaching: catch the exception early

Weekly check-in triage matters because the best coaches do not wait for a full collapse before they intervene. They look for exceptions to the expected pattern.

That is especially true in bodybuilding prep and offseason phases, where the plan often assumes a certain amount of stability. A good check-in should quickly surface:

• appetite changes,
• fatigue changes,
• blood sugar changes,
• dose-response changes,
• adherence breaks,
• and whether the current lever is still the right one.

The KB gives us a useful contrast. Justin’s off-season philosophy is to teach the body to digest and assimilate a massive amount of clean food, building tolerance over time so that more food can be handled without runaway weight gain. He frames that as beneficial for muscle growth, metabolism, and future contest prep. That is a long-horizon objective.

But weekly check-in triage is a short-horizon control problem. If the weekly report shows the athlete is no longer behaving like the model assumed, the coach cannot just keep steering toward the original destination and hope the details sort themselves out. The plan has to match the current state.

This is where AI coaching can actually help, if it is built for decision quality rather than content generation. A useful system should not only summarize “the client is hungrier” or “sleep is down.” It should ask: is that enough to change food, cardio, insulin timing, stimulant use, meal timing, or nothing? And it should know when the answer is “nothing” because the signal is noisy.

Decision quality means knowing what to ignore

Not every check-in deserves a change. That is the part coaches often get wrong when they are trying to be responsive.

If the system reacts to every minor swing, it becomes unstable. If it ignores the wrong thing, it becomes lazy. The art is knowing what is signal and what is noise.

In the KB, Justin repeatedly makes that distinction. He is willing to pause or reduce a retatrutide dose if appetite suppression and fatigue start working against the actual phase objective. He is not committed to the tool for its own sake. He is committed to what the phase needs. Likewise, when insulin sensitivity changes on a high day, he adjusts the dose instead of pretending the old dose should still work.

That is the coaching equivalent of a control loop with a threshold. The weekly check-in exists to detect threshold crossings:

• appetite falls too far;
• fatigue rises beyond normal;
• insulin response changes;
• food tolerance changes;
• adherence starts slipping because the plan has become awkward.

If the threshold is crossed, act. If not, hold.

That sounds basic, but it is exactly where AI systems get sloppy. They over-weight language, under-weight context, and confuse being detailed with being accurate. A client can write three paragraphs and still miss the one line that matters. A coach can respond with ten bullet points and still miss the lever that needs moving.

What the check-in should ask, every time

If you are building or using AI coaching for physique athletes, the weekly check-in should be structured around triage rather than journaling. The check-in should make it hard to miss the following:

1) What changed since last week?

Not “how was the week?” but what specifically moved: appetite, bodyweight trend, hunger timing, training performance, GI tolerance, sleep, mood, blood sugar, or dose response.

2) Is the change phase-relevant?

A small appetite drop during a mild gain phase may be annoying but tolerable. The same appetite drop during a push to add food may be a problem. Context matters.

3) Does the lever still match the goal?

The KB’s retatrutide exchange makes this point cleanly: if a drug is suppressing appetite in a way that helps one phase and hurts another, the decision is phase-specific, not ideology-specific.

4) Is the next action reversible?

Justin’s advice pattern is often to make the smallest reasonable change and then observe. Reduce the dose. Bring meals closer. Watch BG over time. Hold off until prep. This is better than “fixing” everything at once and losing the ability to interpret the result.

5) Did the athlete actually report the important thing?

This is where human coaching still matters, and where AI can assist by structuring the prompt. Many athletes report symptoms but not the decision variable. A coach needs the system to pull the important variable into view.

The falsifiable standard for AI coaching

Here is the standard I would use: if an AI coach cannot reliably identify when a weekly check-in requires a change versus when it requires patience, it is not a coaching tool yet. It is a writing tool.

That is a strict test, but it is fair. Weekly coaching is not about sounding insightful after the fact. It is about making the next decision less wrong.

The KB shows what that looks like in practice. When appetite suppression from retatrutide is strong and fatigue is present, Justin does not romanticize the effect. When insulin sensitivity changes enough to affect meal timing, he reduces the dose. When phase goals shift, he is willing to pause or adjust. The common thread is not rigid adherence to a template. It is correct triage.

That is the practical promise of AI in fitness coaching: not endless personalization, but better first-pass decisions. If the system can surface the right exception, name the right mechanism, and recommend the smallest reversible action, it earns its place. If it cannot do that on a weekly check-in, no amount of polished language will save it.

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
• raw/_consumed/2026-05-26/troponiniq_kb.md