Weekly Check-In Triage and the Insulin Sensitivity Signal

Weekly Check-In Triage and the Insulin Sensitivity Signal

Why better AI coaching will come from fewer, sharper decisions — not more chat

Justin Harris’s check-in handling in the Rory Lazowski and Joe Webb logs is built around one mechanism: sensitivity drift. When Joe’s usual insulin dose started dipping blood sugar sooner than expected on a high day, the response was not a philosophical debate about protocol purity; it was a dose reduction and a meal-timing adjustment on the next check-in. That is the cleanest case for AI coaching triage I can point to: the best weekly check-ins do not optimize everything, they identify the one variable that actually changed and act on it fast. If AI coaching is going to be useful, its value will show up first in weekly triage quality, not in endless feedback volume.

The real job of a check-in is to separate signal from noise

Most athletes send more data than they can interpret. Weight, pumps, hunger, sleep, steps, GI status, training performance, adherence, mood, and sometimes glucose or insulin response all show up in the same message. The failure mode is obvious: coaches get buried in details and answer the wrong question. The better pattern is to triage.

In the Joe Webb exchange, the useful signal was straightforward. He reported that on his high day, the same insulin dose as the prior week brought blood sugar down more noticeably, to the point that he had to move meal 2 up by 30 minutes. After cutting 1 IU, the issue still showed up on the third shot, so the next step was to reduce further. That sequence matters because it shows the decision rule: when a variable changes, do not cling to the old dose just because it worked last week. Adjust to the current response.

That is not a glamorous coaching insight. It is a triage insight. Weekly check-ins should answer three questions, in order:

1. What changed?
2. Is the change performance-relevant?
3. What is the smallest correct adjustment?

If you skip question one, you end up making noise-based changes. If you skip question three, you overreact.

Decision quality beats commentary volume

The Rory Lazowski thread shows another useful piece of triage: appetite suppression is only useful when it helps the current phase. Rory reported that 2 mg of retatrutide wiped out appetite even on low-carb days and brought more fatigue than usual. Justin’s response was not to chase the novelty of a strong appetite effect. It was to notice the phase mismatch and suggest running it while leaning out, when that appetite suppression was an advantage, rather than forcing a gain phase to work around it.

That is the core coaching move: match the tool to the phase.

A lot of AI fitness coaching marketing implies the model should be able to answer everything. In practice, the highest-value use case is narrower: interpret a short weekly check-in, identify the most phase-relevant issue, and recommend one adjustment that preserves the rest of the plan. The better the triage, the less the athlete has to guess whether the coach is reacting to a real problem or just chatting.

Here’s the practical implication for coaches using AI: the system should not reward long replies. It should reward correct prioritization. A five-line response that changes the right variable is better than a 500-word essay that leaves the athlete unsure what to do tomorrow.

The check-in questions that actually matter

A weekly check-in only needs enough structure to surface decision-relevant changes. In bodybuilding terms, the questions are simple:

• Did body weight trend the way it should?
• Did hunger, fatigue, or digestion change enough to affect adherence?
• Did training performance shift in a way that points to recovery, fuel, or load management?
• Did any support tool produce a new response?
• Did the plan become harder to execute than last week?

That last question is underrated. A plan can be “working” and still be heading toward noncompliance. Justin’s own wording across the corpus is repeatedly practical on this point: if the plan is creating a problem, the first move is to decide whether the problem is phase-appropriate or a sign that the current dose, meal spacing, or food volume has crossed a threshold.

This is also where AI can be genuinely useful. Not by pretending to feel what the athlete feels, but by mapping the reported change to the smallest likely decision tree. Appetite down? Is that actually helping the current phase. Blood sugar behaving differently? Is the old dose still valid. Food volume getting too high? Is the issue the macro target, the food choice, or the phase itself.

The failure mode is overconfidence in stale data

The most common bad check-in decision is to treat last week’s response as if it were still true. Joe’s insulin example is the cleanest illustration. Insulin sensitivity was not static; the same dose produced a different effect. Once that happens, holding the old dose because it “worked before” is just stale-data coaching.

That matters for AI because models are especially vulnerable to overgeneralizing from previous patterns. If the athlete says, “same dose, different blood sugar,” the system has to privilege the latest response over the old routine. If the athlete says, “no appetite, a bit more fatigue,” the system has to ask whether reduced intake is a feature or a bug in the current phase. If the athlete says, “this is helping, but it’s changing how I can execute,” the system has to distinguish helpful pressure from unhelpful friction.

The skill is not prediction theater. It is disciplined reassessment.

Why this should change how coaches build their weekly review

If I were building a check-in workflow for coaches, I would not start with more AI-generated text. I would start with triage logic:

• flag the biggest change first;
• compare it to the phase goal;
• choose one action that fixes the highest-leverage problem;
• leave everything else alone unless the new data forces a change.

That is exactly what the best parts of the KB examples do. The response to the insulin issue was not “change everything.” It was “the current dose is too much; reduce it further.” The response to the appetite suppression issue was not “this tool is always good” or “always bad.” It was “it may be useful in a leaning phase; I don’t like forcing appetite lower if it fights the current goal.”

This is why weekly check-ins are the natural home for AI in coaching. Check-ins already compress the week into a small decision set. AI adds value only if it improves the compression.

The falsifiable thesis

If AI fitness coaching is going to outperform mediocre human coaching, the first place it has to win is weekly triage: correctly identifying the one or two changes that matter, matching them to the current phase, and producing a single actionable adjustment. If it cannot do that reliably, more data and more language will only make the coaching worse. That is the thesis, and it is falsifiable in the most practical way possible: watch whether the weekly check-in leads to a better decision tomorrow.

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