How the coaching works
The conversation is the heart of the coaching.
Coach AMA is an AI cycling coach. The coaching happens in an ongoing Discord conversation: you say what is happening, the coach asks and responds, and the plan changes when that conversation gives a reason to change it. Training data and endurance science make the conversation better informed; they do not replace it.
A ride file can show power, heart rate and duration. It cannot know that work was overwhelming, your knee felt wrong or you lost confidence on a descent. You have to say that.
The conversation
The plan is built in dialogue, not handed down.
A useful coaching exchange is more than reporting a score. The coach has to understand what happened, test that understanding against the data, explain the trade-offs and return to the result later.
Ask
Start with the athlete's question, experience or concern. Ask for the missing detail instead of inventing it.
Listen
Treat how the effort felt, what life allows and what the athlete wants as evidence—not decoration around the data.
Reconcile
Compare the conversation with the ride files, training history, recovery trends, goal and long-term direction.
Agree and follow up
Explain the recommendation and its uncertainty. The athlete keeps the final say; the coach remembers the decision and checks what happened.
The conversation is not a feedback form attached to the coaching. It is the coaching. The data is there to make that conversation more specific and less forgetful.
Training principles
A starting structure, not an answer in advance.
The coach begins with evidence-informed defaults, then discusses and adjusts them for the event, training age, available time, preferences and observed response. No intensity distribution wins everywhere, and a rule that repeatedly does not fit the athlete should not survive just because it appears in a reference.
The 80/20 idea is a backbone, not a quota for every day or a promise that one split is universally best.
- 01
Base
A pyramidal base with mostly easy volume and time-efficient work when needed.
- 02
Build
More threshold, VO₂ and event-specific work without adding endless hard days.
- 03
Peak
Sharper race demands, less noise and growing freshness.
- 04
Race
Tactics, fueling and execution—fitness is already built.
- 05
Transition
Planned mental and physical recovery before the next build.
Easy days stay easy
Most riders need about two quality sessions per week. Extra volume should usually be easy before another hard day is added.
Progress one thing at a time
Add a rep, extend the work, shorten recovery or raise the target—not all four in the same week.
Recovery is part of the program
Deloads, tapering and post-season transition are written into the plan; an empty calendar is not the same as planned rest.
Specificity grows toward the event
The closer the race, the more the important sessions should resemble its duration, terrain, intensity and tactical demands.
Conversation and data
Data supports the conversation. It does not replace it.
The athlete's account comes first because it supplies meaning, constraints and consent. The other signals help the coach question, confirm or qualify that account. None is a verdict on its own.
- The athlete's account
- What it helps answerWhat the session and week felt like, what life can hold, what matters now and why.
- What it cannot proveThe coach can only respond to context the athlete shares; dialogue still needs honesty and follow-up.
- CTL, ATL & Form (TSB)
- What it helps answerA compact model of longer-term load, short-term load and their balance.
- What it cannot proveUseful planning heuristics—not direct measurements of fitness, fatigue or readiness.
- Workout execution & RPE
- What it helps answerWhether the intended intervals were completed and how costly the session felt.
- What it cannot provePower without context misses heat, terrain, fueling, illness and life stress.
- HRV, resting HR & sleep
- What it helps answerTrends that may flag poor recovery before a hard day.
- What it cannot proveOne unusual morning is noisy. HRV-guided planning has, at best, a small group-level performance edge.
- Training history & power profile
- What it helps answerWhat the athlete has actually tolerated, where the strengths sit and which efforts are current.
- What it cannot proveA best-effort curve underestimates ability when the athlete has not recently gone all-out.
- Event, terrain & conditions
- What it helps answerWhich fitness, skills, pacing, fueling and heat preparation actually matter.
- What it cannot proveSpecificity guides the plan; it cannot make race-day uncertainty disappear.
Limits of the system
The coach can reason, ask and remember. It can still be wrong.
Coach AMA is an AI, not a person and not a physiological oracle. It can misunderstand incomplete language, overvalue a noisy signal or apply a reasonable general rule badly to one athlete. Its reasoning should be explained and open to correction.
What Coach AMA is
- An AI coach, not a human coach; this should always be clear.
- An ongoing conversation that connects the athlete's account with training history, goals and constraints.
- A periodized training plan that can change without losing its longer-term direction.
- A system that records decisions, preferences and observed responses so later conversations have context.
What Coach AMA is not
- A mind reader. It does not know about pain, stress, fear, motivation or a changed schedule unless the athlete says so.
- A proprietary physiological model trained to predict your exact response.
- A medical professional or a replacement for diagnosis, treatment or an in-person expert.
- A system where a readiness colour, AI confidence or one bad workout automatically decides the week.
Foundation
Relatively strong
Large easy-training volume, limited hard-session frequency, progressive overload, tapering and event specificity.
Applied rule
Useful, not a law
CTL/ATL/TSB ranges, ramp-rate limits, decoupling cutoffs and many weekly-structure rules.
Emerging
Test carefully
Durability protocols, concentrated block designs, heat-transfer claims and other promising but context-sensitive tools.
Representative evidence
The training principles have sources.
These papers are representative anchors for the training ideas above. They do not validate the AI coach as a whole, prove that the science is settled or tell one athlete exactly how they will respond.
- Open study ↗
Seiler & Kjerland, 2006
Observed intensity distribution in well-trained endurance athletes.
- Open study ↗
Bayesian network meta-analysis, 2026
No training-intensity distribution showed a definite universal advantage across outcomes.
- Open study ↗
Rønnestad, Hansen & Ellefsen, 2014
Block periodization of high-intensity intervals in trained cyclists.
- Open study ↗
Wang et al., 2023
Systematic review and meta-analysis of tapering in endurance athletes.
- Open study ↗
Foster et al., 2001
Session RPE as a practical way to quantify internal training load.
- Open study ↗
Manresa-Rocamora et al., 2021
Systematic review and meta-analysis of HRV-guided endurance training.
- Open study ↗
Poole et al., 2016
Critical power as a practical physiological fatigue threshold.
Science supplies starting points and boundaries. The ongoing conversation and the athlete's observed response decide whether those starting points fit this person, in this week, for this goal.
The principle
Without the conversation, it is only a training plan.
The athlete says what the data cannot know. The coach brings structure, evidence, memory and questions. Coaching happens when those two sides keep talking, test decisions in real training and return honestly to what happened.