Skip to content
OpenAthlete

Hybrid coaching: how to coach more athletes with AI

Measure where your coaching hours go, hand the clerical part to software and AI drafts, and keep your time for judgment. Plus a way to price it.

OpenAthlete Team5 min read

Every independent coach hits a ceiling at some point. The number depends on how much contact you sell, but the symptoms are the same: Sunday evenings spent rebuilding weeks because a race moved, the ninth copy of the same threshold session retyped with different paces, and a nagging feeling that the athlete who hasn't uploaded since Thursday deserves a message you haven't had time to write.

"Hybrid coaching" is a grand name for a simple idea: let software and AI do the mechanical part of the job, and keep your hours for the part athletes actually pay you for. The hard bit is drawing the line in the right place. This article is about where to draw it, how to price what's left, and what a week looks like once you have.

Find out where your hours go

Before changing tools, measure. For two weeks, note your coaching time in 15-minute blocks under five headings: writing sessions, adjusting plans, reviewing what athletes did, messaging, and admin (invoices, scheduling calls, chasing payments). It's tedious, and it's the only way to stop guessing.

Most coaches who do this find that writing and adjusting sessions eats more time than they thought, and that a lot of it is clerical: turning "5 × 4' at threshold, 2' jog" into paces for twelve different people, moving a block by a week, copying a recovery week from last cycle. That is the part to hand over first.

Start with plain software, not AI

A surprising share of the gain needs no AI at all, just a training platform used properly:

  • A session library. Write "5 × 4' at threshold" once, drag it onto any calendar.
  • Targets as a percentage of each athlete's threshold, VMA, FTP or max heart rate, so the same session gives each athlete their own paces and you never convert by hand again.
  • Copy a week, shift a block, repeat a session every Tuesday until the race.
  • Activities imported from the watch and compared with the plan, so "did they do it?" takes a glance instead of a message.

If you still write sessions in a spreadsheet or a Word file, this step alone will give you back more evenings than any AI feature.

What AI is good at, and what it isn't

Language models are now decent drafting assistants. Three uses hold up in practice. First, turning a sentence into a structured workout: "15' easy, 6 × 3' at 5 km pace with 90" jog, 10' easy" becomes steps with targets on the athlete's zones. Second, drafting a block from a short brief ("four weeks, base phase, 4 runs a week, long run on Sunday building from 1h15 to 1h45") that you then edit. Third, collecting feedback: a few short questions after each activity, with effort and pain pulled out of the answers, gets you more information than athletes would ever type spontaneously.

The weaknesses are just as real. A model doesn't know that this athlete's Achilles flared the last time you added hill sprints, unless you wrote it down somewhere it can read. It produces plausible, generic plans with complete confidence. It fills gaps instead of asking. Treat every draft like work from a keen junior assistant: useful, faster than starting from a blank page, and never sent without you reading it.

What stays with you

The shape of the season. Whether a niggle means an easy week or a physio. The conversation after a DNF. Saying no to a fourth race in six weeks. Noticing that "fine" in a message from someone who usually writes three paragraphs is not fine. None of this is automatable, and it's exactly what makes an athlete stay with a coach for years. The point of the whole exercise is to have more time for it, not less.

How to price when the tools save you time

Price on contact and responsibility, not on how long the plan took to type. A useful exercise: for each type of client, estimate the minutes per week they cost you, all in. A hypothetical example: plan review 15 minutes, two replies to session comments 10 minutes, a monthly call averaged out at 10 minutes, admin 5 minutes. That's 40 minutes a week, close to three hours a month. Decide what an hour of your work is worth, multiply, then add a margin for the unpaid time (learning, marketing, the athlete who needs a long call before their first marathon).

Then build tiers by what the athlete gets from you, for example:

  • Plan only: a block every four weeks, adjusted once mid-block, replies by message within a few days.
  • Standard: a weekly plan, a weekly written check-in, replies within one working day.
  • Full: everything above plus a monthly call and race-week support.

Write your response time into the offer ("within one working day, not on Sundays"). It protects your evenings and sets an expectation you can meet.

Don't lower your price because an AI drafted a block. The athlete isn't paying for typing; they're paying for your judgment and for someone who notices. What tools change is the choice you face: take more athletes at the same level of attention, or keep the same number and give each one more. Both are valid. Decide on purpose rather than drifting into 35 athletes and the same Sunday evenings.

A week that holds as the roster grows

Monday morning, 30 to 45 minutes. Go through the roster once. Who reported pain? Who hasn't trained in five days? Who is waiting for a reply? Sort athletes into three piles: fine, needs a message, needs a call. Most weeks most athletes are fine, and that's the point of looking at all of them at once.

Tuesday to Thursday. Answer session comments in two batches a day rather than as they arrive. Athletes care that you reply within the time you promised, not that you reply in four minutes.

Friday or Sunday. Plan ahead for those whose block ends. For athletes in a stable phase, draft three or four weeks at once (by hand or with AI), then adjust week by week instead of building each one from scratch.

End of month. A short written review for each athlete: what the month was for, what happened, what's next. It takes ten minutes per person and it's often the message they reread.

Where OpenAthlete fits

OpenAthlete covers the mechanical side described above: a session library with targets in percentage of each athlete's values, bulk move and copy, a coloured compliance strip on every session, and a roster view that lists reported pain, inactivity and unread messages. Planned sessions reach Garmin, COROS and Suunto watches. With AI turned on, you can write a session from a sentence, athletes answer a few questions after each activity, and Claude or ChatGPT can draft a whole block into an athlete's calendar through the MCP server, for you to review. It is free for up to five athletes; Supporter (€5 a month or €50 a year) removes the limit, and a self-hosted instance has none. Details on the coaches page and pricing.

Written by

OpenAthlete Team

OpenAthlete is an open-source training platform built in Grenoble by an independent developer and ultra runner, with feedback from the coaches and athletes who use it. No investors, no ads.

Keep reading

Train with numbers you can check.

Every formula OpenAthlete uses is in its source code, and every feature is free on our servers or yours.

Free forever. No credit card.