Swap a planned exercise for what the gym has

You walk over to the hack squat for your second exercise and the machine has been camped on for twenty minutes. You don't have time to wait. You tell the agent and pick what you'll do instead.

What the agent needs to hear

Behind your sentence, the agent works backwards to two pieces: which planned exercise to swap out, and what to swap it in for. The first is usually obvious from context — the active or next-queued exercise — so « swap it for goblet squat » is enough mid-flow. If you’re swapping something further down the queue (« swap the third exercise, that’s leg curls, for hamstring curls »), name it.

For the substitute, you can name it outright (« for goblet squat ») or hand the decision to the agent (« what can I swap it for? »). When you ask, the agent reads the muscle groups the planned exercise targets and lists two or three substitutes that fit what’s typically available, ordered by closeness of fit.

Letting the agent pick

When the agent suggests substitutes, it looks at three things: the muscle groups the original hit, the movement pattern (squat vs hinge, push vs pull), and the equipment context if you mention it. « Swap hack squat — what can I do? It’s leg day at a hotel gym » will get you bodyweight or dumbbell options, not a leg press alternative.

If the suggestions don’t fit (« I don’t have a kettlebell here »), name the constraint and the agent reshuffles. If none of them feel right, override with your own pick — « actually, I’ll do walking lunges instead » — and the agent attaches that one even if it wasn’t on the suggestion list.

How the swap reads downstream

A swap is not a skip. The system tags the planned exercise as substituted and attaches the work you actually did to the substitute. That distinction matters in your weekly review and routine adherence: a substituted exercise counts toward « the session was completed, with one variation », while a skip counts as « the session was incomplete ».

Set targets typically carry over from the planned exercise — same number of sets, same rep range, weight scaled to the substitute’s loadable range. If the substitute is meaningfully easier or harder than what was planned (goblet squat vs hack squat, very different absolute loads), the agent scales the weight target without changing the rep target. You can override either: « do 4 sets of 12 instead of 4 of 8 » resets the target structure for the substitute.

When the agent gets it wrong

The card showing the swap is where you catch issues. If the agent swapped the wrong exercise (you meant hack squat but it pulled the next one in the queue), correct it: « no, the hack squat — the one I’m doing now ». If the suggested substitute doesn’t match (« that’s a hamstring exercise, I needed something for quads »), call the muscle group and the agent re-picks.

If the agent moved the planned exercise to skipped instead of substituted, push back: « substitute, not skip — I did goblet squat in its place ». The distinction lives in the data and the agent will re-tag.

What makes a clean swap

Three things decide whether this swap reads correctly later: the right planned exercise is the one being replaced (so the routine queue doesn’t end up with the wrong gap), the substitute is a real exercise the system knows (so the work attaches to a tracked line, not a free-text note), and the swap is tagged as substitution, not skip (so adherence reads truth — you completed the session with one variation, not an incomplete session). The system uses these for adherence, weekly review, and routine progression; logging the work as free-form notes or skipping the swap altogether loses the trail. Name the swap as it happens, confirm the substitute, and keep moving.