Start following a diet plan

Sunday evening. You finished last month's maintenance block on Friday and the new cut starts tomorrow. The plan is in your library; you've reviewed it. Now you just tell the agent to switch over so tomorrow's breakfast pulls from the new structure.

What the agent needs to hear

Behind your sentence, the agent works backwards to two pieces: which plan to activate (by name or by descriptor), and — implicitly — what to do with the current engagement, which gets ended automatically so two plans never compete for the daily queue. An optional third piece is the start date, which defaults to now unless you say otherwise.

Name the plan the way it sits in your head: by title (« activate Q2 cut »), by descriptor (« the new cut »), or by what makes it distinct (« the one with the four meals and the post-workout snack »). The agent searches your library for the closest match and shows you a preview card with the plan’s first-day meal queue and macro targets — then waits for a nod before flipping the switch.

What activation actually changes

A plan activation has narrow scope: it changes which plan the next meal log pulls from, and nothing else. Past meal logs stay attached to whichever plan was active when they happened — last month’s maintenance history doesn’t migrate, doesn’t get rewritten, doesn’t get re-scored against the new targets. The activation is forward-looking.

The previous plan’s engagement moves to ended, with its summary captured — adherence, streaks, weekly deltas, all preserved on that engagement record. If you want to cycle back to the old plan later, that’s automatic: ended engagements stay in your history, and re-activating spins up a fresh engagement against the same template.

When the plan isn’t quite ready

Plans can be in draft state — no meals filled in, partial week, missing macro targets on some days. Activating a draft is allowed, but the agent will flag what’s incomplete: « plan has only week 1 filled in; activate anyway? ». If you say yes, the plan activates and the empty days behave as off-plan — free-form logging, no adherence percentage on those days. If you say no, activation is cancelled and the current engagement keeps running.

The same flag fires for plans with no scheduled days, or no items in some meals. None of these block activation; they just warn so you know what’s coming on day 3 when there’s nothing queued.

When the agent gets it wrong

Watch the preview before confirming. If the agent picked the wrong plan (« the old maintenance », instead of the new cut you drafted last week), name the difference more sharply: « no, the new cut, not the old maintenance — the one with 2,400 kcal ». The card swaps to the correct one.

If the agent activated without showing a preview, push back: « wait, show me first — I want to confirm the day-1 queue before committing ». It rolls back and renders the card. Same mechanic for a wrong start date: « Monday, not today » shifts the activation to the start of the week without touching anything else.

What makes a clean activation

Three things decide whether this switch lands cleanly: the right plan is the one now active (so tomorrow’s queue matches what you mean to follow), the previous engagement was ended cleanly, not deleted (so its history stays attached and you can cycle back later), and the draft state is acknowledged if applicable (so you’re not surprised when week 3 has no meals queued). The system uses the active plan for the daily meal queue, adherence scoring, and weekly review; activating the wrong one quietly anchors every future log to the wrong template. Pick the plan the way you’d name it out loud, confirm the day-1 queue, then go.