Training Feedback Loops: 4 Signal Types from AI Coaching Logs
Why better coaching tech should shorten the time between rep, note, and decision—not just count more data
Training Feedback Loops: 4 Signal Types from AI Coaching Logs
Why better coaching tech should shorten the time between rep, note, and decision—not just count more data
The clearest coaching evidence in the Kahunas corpus is not a flashy algorithmic win; it is a recurring decision rule: when feedback is timely and specific, coach and athlete can change the plan within 24 hours, instead of waiting a week for confusion to harden into bad training. That is a latency problem, not a motivation problem, and the mechanism is simple: feedback loops work when they reduce time-to-decision. The thesis is falsifiable: AI fitness coaching is only useful to the extent that it compresses the gap between performance signal and action; if it only adds more check-ins, it adds noise, not coaching.
What the logs actually reward
The corpus repeatedly shows Justin Harris working from a narrow set of training signals: scale trend, look, fatigue, depletion, stubborn body areas, sleep regularity, and readiness to execute. The useful part is not that he has more data than a spreadsheet. It is that he turns those inputs into a decision hierarchy.
One example is brutally practical: an athlete reports being down 2.1 lb for the week, looks more depleted, and still feels they have stubborn lower-back fat. The response is not “trust the process” in the abstract. It is: yes, the lower back is the stubborn area, and the plan should be to get as lean as possible so the rebound can be pushed harder later. The underlying tradeoff is between deeper depletion now and a bigger rebound later. That is a classic feedback-loop move: use the present week’s signal to choose which constraint matters more.
Another example from the same exchange makes the mechanism even clearer. Justin notes that there is a line where pushing depletion further creates more total weight change from end of diet to end of rebound, but not necessarily a bigger end rebound size. In plain English: more loss is not always more progress. Past a point, you are regaining fullness you lost from dieting rather than adding new fullness. That distinction matters because it changes the decision target. If your feedback loop only watches scale loss, it will overvalue unnecessary depletion. If it watches the rebound objective, it will stop at the right time.
The best signal is often not the most measurable one
AI coaching tools love measurable inputs, but the best training decisions in the corpus often come from messy, human signals that are still structured enough to act on.
A client says they average about 8:30 of sleep by WHOOP, have no evidence of apnea, and wake feeling well rested. That is not a green light to do nothing forever. It is a specific stability signal: sleep is not the bottleneck right now. In a feedback system, that means the coach can stop spending decision capital on sleep and allocate it to the actual bottleneck, whether that is recovery from training, exercise selection, or conditioning work.
The same pattern appears in the meal and stimulant example. A client reports a black coffee with meal 1 at 150 mg caffeine and a white Monster with meal 2 at another 150 mg. The point is not the caffeine itself; the point is that the athlete surfaces a detail that changes the interpretation of recovery, appetite, and daytime energy. Good coaching loops do not require perfect data. They require the athlete to tell the truth about the inputs that might alter the decision.
That is where AI can help and where it can fail. It helps when it structures the conversation around the few signals that matter. It fails when it turns everything into a checkbox with equal weight. If every variable gets logged but no variable gets prioritized, the loop becomes administrative instead of adaptive.
Fast feedback beats broad feedback
The Kirtlan Lewis exchange shows the operational advantage of compressed feedback. Justin tells the client he is reading Q&A and will have a full plan out within 24 hours. That is not a glamorous claim. It is better than glamour. In coaching, speed matters because training plans are not static documents; they are hypotheses that decay if the athlete keeps training on an outdated assumption.
This is the part AI coaching platforms should care about most. They can reduce latency in three ways:
- Capture the signal sooner. If the athlete can send a short check-in after training, the coach sees the change before it becomes a pattern.
- Classify the signal better. Is this fat loss, depletion, sleep stability, or just noise from a weird week?
- Return a decision, not a dashboard. The output should be a change in training, nutrition timing, exercise order, or recovery emphasis—not another graph to admire.
The best example of that decision-first approach is in the TroponinIQ framing itself. The platform is described as a coaching ecosystem: AI, structured lecture courses, progress tracking tools, and supplement integration. That mix matters because training feedback loops are not only about AI chat. They are about whether the system can connect what happened in the gym to what should happen next.
Decision tradeoffs are the point, not a bug
A lot of coaches and AI products sell certainty. The logs show something better: explicit tradeoffs.
Justin Harris repeatedly makes decisions under constraint. In one message, he says he does not like prepping hard for longer than 16 weeks and would not push too much at that point, while still trusting the athlete’s understanding of their body. That is a useful model for AI coaching because it admits that there is no universal “more is better.” There is only the question of which constraint is active now.
That is true in training performance feedback too. If the athlete is getting stronger but flattening out, the decision is different than if the athlete is losing bodyweight but not improving readiness. If the athlete reports stable sleep and good execution, the next move is not to chase recovery metrics for their own sake. If the athlete reports a stubborn body area and visible depletion, the move may be to keep pushing, not to soften the plan prematurely.
The tradeoff framework matters because it prevents a common AI failure mode: averaging together all available signals until the recommendation becomes bland. A system that says “sleep looks good, weight is down, and training is okay” but cannot tell you whether to push, hold, or pull back has not coached anything.
What coaches should actually build
If you are building AI coaching for physique athletes, start with the feedback loop, not the interface.
A useful system should answer four questions quickly:
- What changed since last check-in?
- Which signal is dominant right now?
- What tradeoff is being accepted?
- What changes today because of it?
That structure fits the corpus better than generic “how did the week go?” prompts. It also respects the reality that athletes are not machines with a single output. They are systems with competing constraints: recovery, adherence, conditioning, fullness, and schedule.
The practical takeaway for coaches is simple. AI fitness coaching is most valuable when it shortens the path from performance signal to training decision. If the tool cannot help you decide whether to push depletion, hold calories, adjust workload, or ignore a stable metric, then it is just a prettier logbook. The strongest systems are not the ones with the most inputs; they are the ones that turn the right input into the next decision fastest.
Sources Used
raw/_consumed/2026-05-26/troponiniq_kb.mdraw/_consumed/2026-05-31/kahunas-export/2026-05-31-w13-18m/clients/skip_hill___members-b7s-_d0oqqz4a1tadlvuztb0y_vzfqlr9c29tpakweg.jsonraw/_consumed/2026-05-31/kahunas-export/2026-05-31-w19-24m/clients/harry_sims___members-m-iwooxrve6qux4pdtrouhwbkhmvrxoandj1tgzve_m.jsonraw/_consumed/2026-05-31/kahunas-export/2026-05-31-w19-24m/transcripts/alex_goracy___members-m1vvgmmbnhazzipv5wqmwwhgjuyotawsompy4kzf6ri.mdraw/_consumed/2026-05-31/kahunas-export/2026-05-31-w7-12m/clients/kirtlan_lewis___members-skrk4_-y65-elwtevumuhfn1jc9euorjgnpflewxzjc.jsonmodules/03-knowledge/kahunas-coaching-deep-peds.md