Client Memory and the 6-Week Recomp

Client Memory and the 6-Week Recomp

Why repeated coaching notes beat clever prompts when the same mistakes keep returning

The strongest signal in the Kahunas coaching archive is blunt: Justin Harris tells Rory Lazowski, “That’s the hardest thing about coaching. So many things are repeated topics, it’s hard to remember who I discussed them with and who I didn’t,” and the client immediately confirms that prior blood-sugar teaching prevented panic when similar issues returned. The mechanism is simple longitudinal memory: when a coach preserves prior context, repeat problems become expected adjustments instead of fresh firefights. That matters because AI fitness coaching will not fail first on lack of language; it will fail on forgetting. The sharp thesis is this: if an AI coach cannot remember a client’s prior mistakes, the most sophisticated feedback loop in the world will still produce repetitive, lower-quality coaching.

The Rory/Joe exchanges show why this is not a theoretical concern. In the Rory thread from late May, the client says he was “100% expecting” blood sugar issues because he had already been taught about them the year before. Justin’s reply is revealing: he explicitly worries about repetition and says he is “hyper paranoid” that clients may feel neglected, so he repeats things “a dozen times just to be safe.” That is not inefficiency; it is an admission that coaching quality is partly a memory problem. A coach can know the right principle and still miss the fact that the client has already heard it, already tried it, already reacted to it, or already needs a different version of it.

The practical lesson for AI is not “be more human.” It is “be more persistent about state.” In a normal chatbot, every check-in starts too close to zero. It can summarize last week, but it often cannot tell whether a note is a one-off or a recurring pattern. In coaching, that difference matters. If a client repeatedly brings up the same issue—blood sugar swings, food timing, appetite suppression, or dose adjustments—the right response is rarely a brand-new explanation. It is usually a remembered interpretation: what happened last time, what changed, what the client already understood, and what still needs to be watched.

Joe Webb’s thread gives a cleaner example of longitudinal memory in action. On November 30, he reports that his same insulin dose as the prior week caused a noticeable dip in blood sugar, so he brought meals closer together and planned to reduce the dose further on the next high day. Justin’s response on December 2 frames the whole situation inside the larger six-week recomp: the client is sick, holding water, and not going to look as lean as he otherwise would, but he has still dropped body fat. That matters because the coach is not reacting to a single check-in in isolation; he is interpreting it against the known history of the plan. The client’s issue is not “new problem, new panic.” It is “known pattern, updated response.”

That is the core use case for memory-aware AI in fitness coaching. A good system should store more than weights, macros, and screenshots. It should store recurring interpretations. Not just that a client “had blood sugar issues,” but that they were expected after a show rebound, that they were already explained last year, that they are less alarming when the client understands them, and that meal timing adjustments solved the immediate day-level problem. Not just that someone felt appetite suppression, but that a specific dose change was associated with reduced hunger and some fatigue, and that the coach was thinking about whether to pause or reduce it when food needs rose. The point is not to make the AI medically clever. The point is to make it coachingly consistent.

The December Rory exchange also shows how context changes the meaning of the same signal. Rory reported that a 2 mg dose of retatrutide led to essentially no appetite, even on low-carb days, and that he felt more fatigued than normal. He also said he wasn’t sure how that would reconcile with gaining phase food intake. Justin’s response was not to treat appetite suppression as an abstract win. He tied it to real coaching tradeoffs: if food is going up, the dose may need to be paused or reduced until prep, and the value of using it in a lean-out phase is easier to judge than using it in a gain phase. Again, the issue is memory plus framing. A coach who remembers only “appetite went down” can miss the operational consequence: the same intervention can be helpful in one phase and annoying in another.

This is where AI systems usually overpromise and underdeliver. They can summarize a week. They can label a trend. They can draft a response in a polished tone. But if they cannot distinguish between first-time confusion and a problem the client has already survived, they will repeat themselves in the worst way: with confidence. That is expensive in coaching because repetition has different meanings. Repeating a reminder can be useful. Repeating an explanation because the system forgot the previous explanation is wasteful. Repeating a recommendation that the client has already tested is worse: it trains the client to ignore the assistant.

A useful memory stack for AI fitness coaching does not need to be fancy. It needs three layers:

1. Event memory — what happened this week: blood sugar dipped, appetite dropped, fatigue rose, meals had to be brought closer together.
2. Interpretive memory — what it meant last time: expected after the show, already taught, not a reason to freak out, tied to the current phase.
3. Action memory — what worked: reduce insulin dose, shift meal timing, hold or reduce the dose when food increases, keep the client from making the same mistake twice.

If an AI coach can preserve those three layers, it can avoid the most common repeat coaching errors: overexplaining a known issue, underreacting to a new one, and giving advice without the phase context that makes it usable. That is the bar. Anything below it is just a very talkative forgetting machine.

The deeper point is that memory is not a luxury feature for elite coaching; it is the mechanism that turns advice into a relationship with history. Clients do not experience their plans as isolated data points. They experience them as the same five or six recurring problems in new clothing. Coaches who remember that do better because they waste less time re-litigating the obvious and spend more time on the actual decision. AI coaching should be judged by whether it can do the same.

Sources Used:

• raw/_consumed/2026-05-31/kahunas-export/2026-05-28/corpus.jsonl (via the Rory/Joe transcripts in the consumed export)
• 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/transcripts/joe_webb___members-rksigkykimaxwmo_t4_e8nwvbtc2j0etleutkyysads.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/clients/joe_webb___members-rksigkykimaxwmo_t4_e8nwvbtc2j0etleutkyysads.json
• raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w19-24m/transcripts/rory_lazowski___members-c5balaovjbdoeefqmfuqdhh2tbpmfdu16lnf0tnrtmw.md

Sources Used

• /Users/justinharris/TroponinIQ/kb/supertrop/raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w13-18m/clients/rory_lazowski___members-c5balaovjbdoeefqmfuqdhh2tbpmfdu16lnf0tnrtmw.json
• /Users/justinharris/TroponinIQ/kb/supertrop/raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w13-18m/transcripts/rory_lazowski___members-c5balaovjbdoeefqmfuqdhh2tbpmfdu16lnf0tnrtmw.md
• /Users/justinharris/TroponinIQ/kb/supertrop/raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w13-18m/clients/joe_webb___members-rksigkykimaxwmo_t4_e8nwvbtc2j0etleutkyysads.json
• /Users/justinharris/TroponinIQ/kb/supertrop/raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w13-18m/transcripts/joe_webb___members-rksigkykimaxwmo_t4_e8nwvbtc2j0etleutkyysads.md
• /Users/justinharris/TroponinIQ/kb/supertrop/raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w19-24m/transcripts/rory_lazowski___members-c5balaovjbdoeefqmfuqdhh2tbpmfdu16lnf0tnrtmw.md