Weekly Check-In Triage: 3 Decisions That Improve AI Coaching

Weekly Check-In Triage: 3 Decisions That Improve AI Coaching

Decision quality beats more data when the check-in is the bottleneck.

Justin Harris’s coaching examples show a recurring pattern: when blood sugar worsens on growth hormone, he drops the dose and watches a month or two; when retatrutide crushes appetite and adds fatigue, he pauses or reduces it; when high-day insulin needs change, he cuts the dose and shortens the gap to the next meal. The mechanism is simple appetite- and glucose-driven feedback, but the thesis is sharper: better weekly check-ins are not about collecting more metrics, they are about making one correct triage decision before the next week compounds the error.

Weekly check-ins are decision nodes, not diary entries

A lot of AI coaching gets marketed like it is a magical memory system. Log everything. Track more. Let the model see the whole picture. That sounds useful, but in practice most weekly check-ins are not data problems. They are triage problems.

The coach has to answer a few questions fast:

1. Is this trend acceptable, or is it the start of a bad slope?
2. Is the current tool helping enough to keep it, or are the side effects now part of the cost?
3. Is the issue solving itself with timing and dose adjustment, or does it need a harder stop?

That is exactly how the best examples in the KB read. Justin does not treat every update as equally important. He looks for the first sign that the current plan has crossed from productive into counterproductive, then he changes the smallest thing that fixes the problem.

1) Appetite suppression is not automatically a win

In the Rory Lazowski exchange, retatrutide at 2 mg sharply reduced appetite, including on low-carb days, and brought more fatigue than usual. Justin’s response was not “great, lower appetite means better compliance.” He said he was trying it himself to get first-hand experience, but he was not convinced by the idea of forcing appetite lower. Then he made the actual coaching call: keep leaning out while it is easy, and use that period to learn how the drug behaves before deciding what it means for gaining.

That is a useful triage principle for AI coaching: a tool that changes hunger is not inherently good or bad; the decision depends on the phase objective. If the current block is pushing food intake up, then heavy appetite suppression can become a bottleneck. If the current block is a mild cut or a phase where you want to make eating easier, the same effect may be useful. The key is that the check-in should not ask, “Did the drug do something?” It should ask, “Did the drug help this block’s job?”

AI systems tend to overvalue dramatic signal. “No appetite whatsoever” looks like a powerful result, and it is. But power is not the same as fit. For a weekly check-in, the question is not whether the effect is strong; it is whether the effect matches the current task.

A practical triage rule follows from that:

• If appetite is down and food targets are already hard to hit, the problem is not motivation. It is interference.
• If appetite is down during a phase where intake is being deliberately pulled back, the effect may be useful.
• If fatigue shows up at the same time as appetite suppression, the check-in should not wait for a trend to become obvious.

That is decision quality: matching the tool to the phase before the week gets away from you.

2) Blood sugar changes are a dose problem before they are a philosophy problem

In the Joe Webb check-in, the issue was not abstract “insulin sensitivity.” It was concrete: the same insulin dose that worked last week started dipping blood sugar more noticeably on a high day, so meals had to be brought closer together and the dose reduced by 1 IU. Justin did not overcomplicate it. He answered the immediate mismatch with dose reduction and timing adjustment.

That matters because AI coaching often turns a simple execution problem into a long diagnostic essay. Weekly check-ins are where that fails. If the same input now causes a stronger response, the first move is not to invent a new theory. The first move is to reduce the dose and retest under the next week’s conditions.

This is the part many dashboards miss. A better model does not mean a more elaborate explanation; it means a faster recognition of when yesterday’s settings no longer fit today’s response. In the Joe example, the signal was not a lab report or a broad health trend. It was a real-time meal-timing problem during a high day. That is exactly the kind of issue weekly coaching should catch.

The decision chain is clean:

• same dose,
• stronger blood sugar drop,
• meal timing has to be pulled forward,
• reduce the dose further next high day.

No drama. No overreach. Just a better next rep.

3) “How it feels” is useful only if it changes the next action

Justin’s voice examples make another point that AI coaches should steal: he is comfortable naming tradeoffs without pretending they are symmetrical. On growth hormone, he says there is a response curve where the upside eventually reverses, the crossover point varies by person, and higher year-round doses push most people down the wrong side of that curve. If the numbers have clearly worsened, he drops the dose and watches BG over a month or two.

That is not a blanket anti-GH stance. It is a triage stance. If a weekly check-in shows the current dose is pushing a bad response, the dose is too much for that person at that time. The answer is not to keep the same dose because the drug has a reputation for being useful. It is to accept that the response curve is real and move back toward the productive part of it.

This is a valuable pattern for AI coaching more broadly: treat the check-in as a control system, not a transcript archive. The job is to decide whether the current setting still belongs in the operating range.

What good triage looks like in an AI coaching stack

If you are building or using AI coaching, weekly check-ins should be structured to force action, not just status reporting.

A strong check-in should identify:

• the phase goal,
• the most sensitive variable,
• the first sign the plan is drifting,
• the smallest change that restores fit.

That means the model should be trained to ask narrower questions:

• Are meals still getting completed without forcing?
• Did the current dose create a new timing problem?
• Did fatigue rise with the intervention?
• Is the same setting still appropriate for this phase?

The KB sources keep pointing to the same operational truth: most coaching errors are not fixed by more information. They are fixed by earlier recognition of mismatch.

The falsifiable thesis

If weekly check-ins improve decision quality, they should produce fewer weeks where the coach keeps the same setting despite a clearly worse response. The Rory example shows appetite suppression plus fatigue leading to a pause-or-reduce call. The Joe example shows a repeated blood sugar dip leading to dose reduction and meal timing changes. The GH example shows a worsening response leading to a drop and a month-or-two watch period. That is the standard.

An AI coaching system that cannot make those calls is not failing because it lacks enough data. It is failing because it cannot triage.

Sources Used

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