Client Memory, 2 Rebounds, and the Repeated Mistake Problem

Client Memory, 2 Rebounds, and the Repeated Mistake Problem

Why longitudinal memory matters more than “smart” check-ins when coaching needs to avoid the same error twice.

Justin Harris says the hardest part of coaching is that “so many things are repeated topics,” and he repeats things “a dozen times just to be safe”; in Rory Lazowski’s case, that repetition fit a very specific mechanism: expected blood-sugar drift after a show rebound. The sharp thesis is simple and falsifiable: AI fitness coaching will not become meaningfully better because it can answer faster; it will get better only when it can remember what was already explained, what was already expected, and what has already failed.

The coaching problem is not information scarcity

In the Rory Lazowski thread from late May 2024, Justin explicitly connects prior education to better execution. Rory says the blood sugar issues were “100% expected,” because they had already discussed them when they started together the day after the show. Justin’s response is the important part: he notes that the hardest thing about coaching is remembering which topics have already been covered, and he compensates by repeating key points multiple times.

That is not filler. It is a practical workaround for a memory bottleneck. When the same client returns with the same pattern, the coach is not re-discovering the issue from scratch; he is trying to avoid re-litigating it. The useful coaching question is not “Can I explain this better today?” It is “Will I remember that I already explained this last rebound?”

That distinction matters because repeated mistakes usually do not come from ignorance alone. They come from a broken longitudinal record. A coach who forgets that a client already struggled with rebound blood sugar, hunger, fatigue, or dosing tolerance will reintroduce avoidable friction. A coach who remembers can move straight to the next decision.

Rory’s two rebounds show why context beats raw recall

The Rory exchanges are useful because they are not abstract. They span two different post-show periods and show the exact kind of pattern a coach should want an AI system to retain.

In the May 2024 thread, Justin says Rory had already been taught about the blood sugar issue the previous year, and that last year was different because Rory “had no idea” why it was happening. This year, by contrast, Rory says it came as no surprise. That is the win: the same physiological pattern did not become magically benign; it became operationally manageable because the client remembered the explanation and did not panic.

That kind of memory does not merely make a conversation more pleasant. It changes the coaching loop. If the issue is expected, the coach can avoid over-correcting. If the issue is surprising, the coach may have to spend the next several messages rebuilding trust, confidence, and adherence.

Now compare that with the December 2024 exchange in the same client history. The subject shifts to retatrutide. Rory reports that 2 mg on Friday produced essentially no appetite, even on low-carb days, unlike semaglutide/Ozempic. He also notes increased fatigue, and says he may pause or reduce the dose if food is being added.

The coaching value here is not the appetite effect itself; it is the record. A coach who remembers the prior pattern can interpret a later appetite suppression report in context instead of treating it like a brand-new breakthrough. The client’s body is already telling a story across time. The coach’s job is to keep that story intact.

The mechanism is longitudinal memory, not chatbot fluency

If an AI coach is only as good as the current prompt, it is just a very polished forgetful assistant. The mechanism that matters is longitudinal memory: the ability to retain prior explanations, prior problem areas, and prior decisions so the next recommendation is conditioned on the real history.

That history is what prevents repeated errors such as:

• revisiting a known issue as if it were new,
• making a change without accounting for an old tolerance pattern,
• missing the difference between “unexpected problem” and “expected but manageable problem,”
• and overestimating how much a client needs to be reassured when the client already understands the pattern.

The Rory threads show all four. In May, the issue was not that blood sugar drift existed; it was whether it was expected. In December, the issue was not whether retatrutide had an appetite effect; it was how that effect should alter the plan.

That is the difference between a coach who remembers and a system that only reacts.

Why AI coaching systems are especially vulnerable here

AI tools are usually evaluated on the wrong axis: response quality in a single turn. Coaches do not live in single turns. They live in recurring patterns across weeks, cuts, rebounds, off-seasons, travel blocks, sickness, appetite changes, and repeated adjustments that only make sense if the prior ones are still available.

When memory is weak, the system keeps re-asking the client to carry the history. That creates three avoidable problems.

First, the client does the remembering. That is a burden, especially when the client is already tracking weight, intake, pumps, training performance, and subjective stress.

Second, the coach looks inattentive even when he is not. Justin’s comment about repeating things “a dozen times just to be safe” is a human workaround for a real reputational hazard: clients notice when they have to re-explain the same thing.

Third, the plan gets noisier. If a coach fails to remember that a blood sugar issue was already discussed, or that appetite suppression showed up after a specific change, then every new recommendation starts with avoidable confusion.

This is where AI can actually help if it is designed for memory rather than novelty. The practical win is not “more insights.” It is fewer repeated coaching mistakes.

What coaches should store, not just what they should say

The best longitudinal notes are not essays. They are durable facts that can be retrieved later. For a fitness coach, that usually means:

• what the client has already been taught,
• what was expected versus unexpected,
• what changed when the issue appeared,
• which interventions were tried,
• and what the client already knows how to do without hand-holding.

The Rory examples make the case cleanly. One thread documents a post-show blood sugar issue that was understood because it had been taught previously. Another records a strong appetite-suppression response at 2 mg retatrutide with some fatigue. If those details are preserved, the next decision can be precise instead of generic.

That is especially important in repeated phases of coaching where the same patterns return under different labels. A rebound is not just “weight gain after a show.” It may be the same old blood sugar conversation showing up in a new context. A later appetite change is not just “the client feels different.” It may be a repeat pattern with a different trigger.

AI systems that do not retain those distinctions will keep rediscovering them at the client’s expense.

The practical thesis for coaches

The take-home is not that AI should replace coaches. It is that memory should be treated as core coaching infrastructure. If the system cannot remember the last rebound, the last explanation, and the last reason a plan was modified, then it will keep making the same mistakes under a new interface.

That is the falsifiable test. Put an AI coaching system in front of a recurring client problem and ask whether it:

1. recognizes the prior discussion,
2. distinguishes expected from unexpected changes,
3. preserves what was already learned,
4. and reduces repeated corrections over time.

If it cannot do those things, it is not doing coaching memory. It is just generating text.

Sources Used

• raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w19-24m/clients/rory_lazowski___members-c5balaovjbdoeefqmfuqdhh2tbpmfdu16lnf0tnrtmw.json
• raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w19-24m/transcripts/rory_lazowski___members-c5balaovjbdoeefqmfuqdhh2tbpmfdu16lnf0tnrtmw.md
• 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