Weekly check-in triage is a decision system, not a report

Weekly check-in triage is a decision system, not a report

Justin Harris’s coaching logs show a simple pattern: when insulin sensitivity changed on a high day, meals were pulled closer together and the insulin dose was reduced; when appetite suppression from 2 mg retatrutide was strong enough to flatten hunger and add fatigue, the first move was to lean out while it was easy and revisit the dose later. The mechanism is straightforward: dose-response plus feedback control. That is the core test for AI fitness coaching. If weekly check-ins do not improve triage—what to change, what to hold, and what to watch next—they are just verbose summaries, and verbose summaries do not coach.

The weekly check-in is where coaching either earns its keep or exposes its weak spots

A lot of AI coaching products want to sell “24/7 access” as if always-on availability were the same thing as better decisions. It is not. The useful unit is not chat volume. It is whether a weekly check-in improves the quality of the next action.

That is why triage matters. A good check-in does three jobs:

1. identifies the dominant constraint,
2. chooses the smallest effective adjustment,
3. sets a short feedback window.

Miss any one of those and the next week becomes noise.

The check-in example from Joe Webb is a clean case. On a high day, the same insulin dose that had worked the prior week now dipped blood sugar enough that meal 2 had to come 30 minutes sooner. The response was not to rewrite the whole setup. The response was to reduce the dose further on the next high day and keep meals closer together. That is what decision quality looks like in practice: matching the intervention to the problem you actually saw, not the problem you imagined.

This is the first lesson for AI coaching. The system has to distinguish between a stable plan and a plan that is now out of date. If the athlete’s response has changed, the coach should not preserve the old decision out of habit. Weekly check-ins exist to catch exactly that sort of drift.

Appetite, fatigue, and the trap of confusing symptom relief with better coaching

The retatrutide exchange is useful because it shows how easy it is to overinterpret a strong signal. Rory Lazowski reported 2 mg and said appetite disappeared, even on low-carb days, with a bit more fatigue than normal. Justin’s reaction was not “great, keep pushing.” It was more restrained: he was trying it himself for first-hand experience, but he did not love the idea of forcing appetite lower. He then said the cleaner move was to run with it while leaning out, because that would make the next phase easier to interpret.

That is the right instinct for weekly triage. A dramatic appetite shift is not automatically a green light for more aggressive gaining. If the current phase is already making food easier to manage, you use that window to get a better baseline. Then you revisit the appetite tool when the context changes.

This is where AI coaches often get sloppy. They can confuse “the athlete feels different” with “the athlete should do more.” But the more reliable move is often to ask: does this change help us make better decisions over the next two to four weeks? If the answer is no, the check-in should not escalate the intervention just because the signal is loud.

What a strong weekly check-in should always return

A useful check-in does not need a heroic amount of prose. It needs the same three outputs every time.

1) What changed?

Not “how was the week?” That is too broad. The question is what changed in the measurable or observable state that matters for the next decision. In the examples above, the important changes were:

• blood sugar dipped sooner on the same insulin dose,
• appetite dropped sharply after 2 mg retatrutide,
• fatigue increased alongside the appetite drop.

Those are not personality notes. They are decision inputs.

2) What is the smallest effective adjustment?

A good coach does not swing for the fences on every check-in. In the high-day insulin example, the adjustment was small and specific: reduce the dose further and bring meals closer together. In the retatrutide example, the response was not to chase the appetite suppression higher; it was to consider pausing or reducing the dose if food was about to increase, while using the current phase to lean out.

That discipline matters because many bad coaching outcomes come from overcorrection. Weekly check-ins are supposed to reduce the size of the error bar, not create a new one.

3) What is the next observation window?

This is the part most systems underuse. A check-in without a time horizon is just a conversation. Joe’s message implicitly set the next high day as the test window. Justin’s retatrutide response implied a short-cycle revisit once the phase context changed.

In practice, that means the coach should always answer: what should be true by next check-in if this change was right? If that answer is unclear, the adjustment is probably too vague.

The hidden skill is knowing when not to change anything

Decision quality is not the same as intervention frequency. Sometimes the correct triage decision is to hold steady.

Justin’s own comments about appetite management show this caution. He said he did not like the idea of forcing appetite lower, and he wanted more first-hand data before acting like he had a settled opinion. That skepticism is useful. It keeps the coach from turning every new tool into a permanent default.

For AI coaching, this is an important design principle: the model should be able to say “hold” with conviction. If every check-in ends in an adjustment, the system is signaling instability, not expertise. Coaches know this instinctively. The weekly report should not feel like a vendor trying to justify its existence.

A practical triage rubric for coaches using AI

If you are using AI to support weekly check-ins, ask whether the output answers these four questions cleanly:

• Is this a real change or just day-to-day noise?
• Is the issue a volume problem, a timing problem, or a context problem?
• What is the lowest-risk adjustment that directly addresses it?
• When do we reassess?

That rubric fits both examples here. Blood sugar drift on a high day is a timing and dose problem. Strong appetite suppression with fatigue is a context problem. The right response in both cases is not maximalism; it is matching the fix to the mechanism.

That is also why AI coaching should be judged on triage quality rather than flair. A model that writes polished weekly summaries but misses the actual decision point is not helping. A model that flags the right change, keeps the change small, and sets a clear recheck date is doing the work.

The falsifiable thesis

If AI fitness coaching improves weekly check-in triage, you should see fewer unnecessary changes, faster correction of genuine drift, and better continuity from week to week. If you do not see those three things, the system is not improving decision quality—it is just producing more text.

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