Building an AI-driven running, cycling and swimming plan
A multisport plan is hard to hold because it balances three loads landing in the same body. That is exactly the kind of trade-off an AI handles well — provided you give it the right constraints. Here is how to go about it, and where it gets things wrong.
Last updated: 2026-08-24
What a plan must settle before it is written
A plan is not a list of sessions: it is a sequence of trade-offs. How many sessions a week, split how across three disciplines, with what progression, and which rest days are non-negotiable.
Those trade-offs depend on you, not on theory. An assistant that does not know them will produce a magazine plan: elegant on paper, unliveable in your week.
The constraints without which nothing holds
Plan quality depends almost entirely on how precisely you describe the starting point.
- The goal and its date: a race format, a distance, a target time, and the weeks remaining.
- Your real level, in numbers: ten-kilometre pace, FTP, time per hundred metres. Feel alone cannot calibrate targets.
- Your actual availability: number of sessions, length of each, impossible days, access to a pool or a trainer.
- Your history: injuries, layoffs, and the sports where you start furthest back.
Those four points fit in ten lines and change the result completely.
Where the AI most often gets it wrong
An assistant has read a great many training plans. That gives it a good sense of classic structures, and three stubborn biases.
- Progressing too fast: with no load reference, it raises volume quicker than a body adapts.
- Forgetting that load accumulates across sports: three hard sessions in three different disciplines are still three hard sessions.
- Too much quality: too many target sessions, not enough easy volume, which is most of the work.
- The forgotten recovery week, when that is precisely what makes you progress.
The best safeguard is to have it review its own plan, asking for weekly load and week-on-week progression. The excesses are obvious as soon as they are quantified.
A plan produced and then criticised by the same assistant is markedly better than one produced in a single pass.
From plan to watch
A plan that stays in a conversation is useless: what matters is that today's session is on your wrist, structured, on the right date. That is the step most tools leave to you.
With a write-capable server, the assistant places each session on the Garmin calendar itself, with the targets specific to each sport. The plan stops being a document and becomes an executable week.
Frequently asked questions
Can an AI replace a coach?
It does replace the mechanical part well: structuring, dosing, planning, adjusting from data. It does not replace an eye on your technique, or judgement about what you are not saying. For a recreational athlete with a clear goal, that covers most of it.
How many weeks should a plan cover?
Twelve to sixteen for a long goal, eight for a short format. Beyond that, precision is illusory: the plan will be rewritten anyway as real life happens.
How do I handle a missed week?
By not making it up. The instinct to compensate for missed sessions is the most common cause of injury. Resuming the plan where it now stands, even if progression slips a week, is almost always right.
Should each discipline have its own plan?
No, and that is the classic mistake: three single-sport plans stacked together produce an unmanageable load. A multisport plan has to be thought of as one fatigue budget split three ways.
Let your assistant write your sessions
Running Workout connects Claude, ChatGPT or Gemini to your Garmin account through the official API. Nothing to install, and no Garmin password to hand over.
See the connection guide