VBT Training Logs: 1 Coach, 1 AI, and the Judgment Gap

VBT Training Logs: 1 Coach, 1 AI, and the Judgment Gap

Coach leverage comes from faster filtering, not outsourcing decisions; the system scales review, but the human still owns the call.

On the AI side, the strongest practical result in the KB is not a flashy “AI learns your athlete” claim: it is that TroponinIQ is built to deliver Justin Harris’s coaching brain 24/7 across 100+ pro athletes and 1,000+ transformations, while the training-science layer still centers on human frameworks like RPE/RIR validation, velocity-based training, and volume landmarks. The mechanism is simple: pattern recognition at scale with a judgment gate. If AI coaching is going to matter in a serious physique roster, its job is to compress review time and surface options, not replace the coach’s call. That makes the falsifiable thesis easy to state: AI increases coach leverage only when it improves triage and consistency without taking over decision authority.

That matters because the real bottleneck in coaching is not information scarcity. It is attention scarcity. A coach can know the broad rules—volume matters, autoregulation matters, load-velocity profiling matters—and still lose leverage when every check-in, every food adjustment, and every training tweak requires the same full cognitive effort. The value of AI in this setting is not that it discovers a new hypertrophy law. It is that it lets one coach handle more athletes without flattening the individual judgment that actually makes the coaching good.

The KB points to a useful division of labor. The research layer emphasizes the sort of tools that can be standardized: RPE and RIR models validated by Helms and Zourdos, load-velocity profiling and velocity-based training associated with Bryan Mann, and the practical hypertrophy literature around sets, frequency, and loading that give coaches a stable baseline. Those tools are useful precisely because they are not mystical. They turn subjective training stress into something a coach can inspect quickly. AI can be very good at that first pass: organizing logs, flagging drift, summarizing trends, and making it easier to spot when a lifter is clearly moving away from the expected response.

But the same KB also shows why the final decision cannot be automated cleanly. Justin’s own coaching voice is skeptical, direct, and mechanism-first. In one example, when an athlete asked about retatrutide, the response was not “always use it” or “never use it.” It was conditional: appetite suppression was real, fatigue was real, and if the plan was to add food, the dose might need to be paused or reduced; if the athlete was leaning out easily, that was the time to run with it and gather better data before scaling food up. That is not a generic AI output. It is a judgment call built on context, timing, and the coach’s willingness to update the plan based on what the body is doing, not what the spreadsheet wants.

That is the standard AI has to meet in coaching: be useful where the decision tree is shallow, stay out of the way where the decision tree is deep. A shallow-tree task looks like this: a lifter reports that the same insulin dose that worked last week is now pushing glucose lower and forcing meal timing to tighten. That is a monitoring problem, not an identity crisis. An AI system can surface the trend, summarize the deviation, and remind the coach that dose-response changed. The coach then decides whether to adjust insulin, move meals, or change the day’s structure. The important point is that the system reduces the time to detect the problem, but the response still belongs to the coach.

That distinction is what gives AI real leverage in a roster. At one athlete, manual review is fine. At 20 or 40, manual review becomes triage. The platform promise, the one that matters, is not “we have an AI that coaches.” It is “we have a system that lets a good coach review more athletes with fewer blind spots.” In practice that means four things. First, the AI should standardize intake so the coach sees comparable data across athletes. Second, it should compress daily notes into decision-relevant summaries. Third, it should highlight deviations from prior response patterns. Fourth, it should preserve the coach’s ability to override the machine without friction.

That last point is where a lot of AI hype falls apart. Coaches do not need a machine that sounds confident. They need a machine that knows when it is merely classifying. The stronger the roster, the more obvious this becomes. A serious physique coach is not just managing workouts; he is managing digestion tolerance, appetite, fatigue, body-comp trajectory, and the timing of when to push or hold. The KB’s off-season language makes that clear: the goal is to build the ability to digest and assimilate large amounts of food over time, not to force a short-term win from one intervention. That kind of plan is a sequence of tradeoffs. AI can speed up the observation layer, but it cannot decide the order of those tradeoffs on its own.

There is also a hidden risk in over-automation: it can make a coach worse at judgment by making every case feel like a template. If the system learns to produce tidy recommendations too quickly, the coach may stop asking whether the underlying assumption actually fits the athlete in front of him. That is especially dangerous in bodybuilding, where “good enough on average” is often not good enough for the person chasing a specific outcome. The right use of AI is not to smooth over differences. It is to expose them faster.

So what should coaches expect from AI in 2026? Not magic. Not replacement. Leverage. If the system saves ten minutes per athlete per week and turns scattered logs into a clear decision queue, that is meaningful. If it helps one coach cover more athletes while staying consistent on the fundamentals, that is real scale. But if it starts making the final call for the coach, it becomes a liability. The machine can handle structure. The coach has to handle judgment.

That is the line worth protecting. AI should make a good coach more available, more consistent, and more scalable. It should not make coaching feel automatic. The moment it does, the roster gets bigger and the thinking gets smaller.

Sources Used

• raw/_consumed/2026-06-02/_GRAS/gras_strategy_training.md
• raw/_consumed/2026-05-26/troponiniq_kb.md
• raw/Justin_TT1.txt
• raw/_consumed/2026-05-31/kahunas-export/2026-05-31-w13-18m/transcripts/rory_lazowski___members-c5balaovjbdoeefqmfuqdhh2tbpmfdu16lnf0tnrtmw.md
• modules/08-voice/kahunas-coaching-deep-voice.md