What AI should and should not do in a coach's workflow

Every coaching tool now claims to have AI. Some of it saves real time. Some of it is a chat window bolted onto the side of an app, and some of it quietly makes decisions a coach would never sign off. The difference matters, because the thing an athlete pays a coach for is judgement, and judgement is exactly what you should not hand over.

This article draws a line through a coach's week: the jobs AI does well, the jobs it should only assist with, and the jobs it should stay out of entirely.

The principle: AI does the typing and the reading, you do the deciding

Look at where a coach's hours actually go. A large share is not coaching. It is typing the same session structure again, reading logs to work out what happened, scrolling a roster to see who has gone quiet, and writing first drafts of messages. That work is necessary, repetitive and mostly mechanical.

The coaching is what sits on top: choosing the emphasis of a block, deciding whether an athlete is ready to progress, knowing when a message needs to be a phone call. That is where your experience earns its fee.

A useful rule: AI should produce drafts and readings; a person should approve anything that changes an athlete's training or reaches the athlete. Everything below follows from that.

Jobs AI does well

Turning your shorthand into structure

You write "5 x 3 @ 85%, rest 3 min" or "6 x 800 m @ 3:30, 2 min jog". Reading that into sets, reps, load, rest and interval stages is a well-defined task with a right answer, and you can check it at a glance. This is the single most useful thing AI does for a coach, because everything else (one-tap logging, watch workouts, progress charts) depends on the session being structure rather than text.

The safeguard is simple: show what was read, keep the coach's original words, and let any field be corrected in one click. See structured prescriptions.

Reading logs so you do not have to

Comparing what was prescribed with what was done, across every set and every athlete, is a reading job. AI is good at summarising it: "hit every rep on the main lift, RPE drifting up on the accessory work, cut the last set of the second exercise short twice". You still decide what that means. You just do not have to dig it out.

Pointing at who needs attention

A roster sorted by "last message" rewards the athletes who talk most. Ranking by evidence (who stopped logging, whose compliance is slipping, who just set a record) is a better starting point, and a short morning summary of it saves the first half hour of the day. On Loadout this is the roster and morning briefing.

Jobs AI should assist with, not own

Drafting a block

A first draft of a twelve-week block, built from your brief and the athlete's own answers about schedule, equipment and injuries, is a good starting point. It turns a blank page into something to react to, and reacting is faster than writing.

But a draft is a draft. It should arrive as something you read, challenge and edit, and nothing should reach the athlete's calendar until you apply it. If a tool writes programmes straight into athletes' calendars without a coach in the loop, it is not assisting you; it is replacing you, badly. More on the review step on the AI programme drafting page.

Suggesting the next load

Given the last few sessions of an exercise, prescribed against actual, and how the athlete has been sleeping and recovering, a suggested next prescription with a stated reason is useful. It is a second opinion that has read every rep. It should come with a confidence and a rationale, and it should change nothing until you accept it. See progression suggestions.

Drafting messages

When five athletes need a check-in, a draft per athlete based on what actually happened ("missed Tuesday and Thursday, squat moved up 5 kg on Saturday") saves time. You edit and send. The athlete hears from you, in your voice, about something specific.

Jobs AI should stay out of

Sending anything on your behalf

An athlete who discovers that the encouraging message came from software will discount every message after it. Drafts are fine. Automatic sending is not.

Changing a programme without approval

Silent changes destroy trust in both directions. The athlete does not know whether the session reflects your judgement, and you do not know what they actually did. If the plan changes, a coach should have approved the change.

Deciding about pain, injury and health

Any time an athlete mentions pain, illness or something a doctor said, a person should read it. Software can flag the message. It should not decide what happens next.

The relationship

Knowing when an athlete needs pushing and when they need a week off for reasons that will never appear in a log is the job. No reading of the data replaces it.

A test for any AI feature

When you evaluate coaching software, run each AI feature through four questions:

1. What does it read? If the underlying programme is free text, the AI is guessing at the numbers. Ask what it works from.
2. What does it change? If the answer is "the athlete's training" or "what the athlete receives", ask where the approval step is.
3. Can you see its reasoning? A suggestion without a reason cannot be checked, so it cannot be trusted.
4. Can you turn it off and still work? Every AI step should have a manual path beside it. If the tool cannot be used without the AI, the AI is a dependency, not an assistant.

Where this leaves the coach

Done properly, AI takes the mechanical hours out of the week (typing structure, reading logs, finding who needs you, drafting the obvious) and leaves more of your time for the parts athletes value: decisions, feedback and the relationship. That is the trade worth making. Anything that goes further is asking you to sell judgement you have not exercised.

FAQ

Will AI replace online coaches?

Not the good ones. It replaces typing, reading and first drafts. Athletes pay coaches for judgement and accountability, and those still need a person.

Is it safe to let AI write training programmes?

As a draft that a qualified coach reviews and edits, yes. Applied directly to athletes without review, no.

Do athletes need to know AI is involved?

Be open about it. Explain that the software drafts and summarises, and that every programme and every message is reviewed by you. Most athletes care that a person is accountable, not which tool did the typing.

See it in a real workflow

Loadout's AI drafts, reads and flags; it never applies a programme or sends a message without you. Try it with two athletes, free: start free.