Meta MuseUndisclosed

A personal trainer that runs itself: it asked about health and gym equipment, posts the next workout on off days and logs the lifts each evening

Zack GabbardX (Twitter) · 2026-09-18

Read original

@Muse is fantastic! Very impressed. @finkd After importing memory, I asked it to build a personal trainer and it went full systems-engineer on it. It First asked me a series of questions related to my health, gym access, limitations etc. Then it built a program based off of my answers. Including checking the gym I go to and analyzing the equipment available. Every morning it checks my calendar. If I am working it stays silent. Off days, it counts this week's sessions and posts the next workout in the rotation. Evenings it checks in on how the lifts went and logs it. Here is the card it produces each morning of a workout.

Business model

Zack Gabbard writes that after importing memory he asked Muse to build him a personal trainer, and it went full systems-engineer on it. The process is specific: it first asked a series of questions about his health, his gym access and his limitations; it built a programme from his answers, including checking the gym he goes to and analysing the equipment available; every morning it checks his calendar and stays silent if he is working, while on off days it counts this week's sessions and posts the next workout in the rotation; in the evening it checks in on how the lifts went and logs them. A card generated each workout morning is attached. The post does not say how many weeks he actually trained, whether any weight went up, whether he was injured, what the card contains, where the data lives, which gym's equipment list was used, or what happens if he does not reply in the evening. What it shows is that the routine was built and runs on a schedule - not that it produced results. All of this is his account; AgentFlow imported no health data and ran nothing.

Training plans rarely fail because the plan is bad; they fail because what am I doing today has to be decided again every day, and undecided means not going. Fixing that decision into a calendar-triggered loop is the replicable part. What it cannot supply is the session itself, or any result: the post reports no weeks trained, no load progression and no injury data, and posting a card is not training.

Workflow

  1. Run the intake once and get the constraints complete: injury history and any doctor's limits, training days per week and session length, the gym and available equipment, current working weights, and movements that are off limits. Ask about anything missing instead of filling in a template.
  2. Build the split from those constraints: a four-to-six-week rotation with sets, rep ranges, target RPE and a substitute for every movement. When equipment is missing, the substitute goes on the card with the reason.
  3. Wire the calendar as a read-only trigger: it decides whether today is a work day or an off day, and never writes to or edits the calendar. The rule is fixed - silent on work days, post on off days, drop one level after two missed sessions rather than skipping a cycle.
  4. Post one card each training morning: movements, sets, reps, last weight, target RPE and substitutes. One screen, no motivational essay, no supplement or course promotion.
  5. Collect the evening log: whether it was done, actual weight and reps, any pain, and a reason for skipped sessions. If there is no reply, do not chase twice; mark the next card as last session unlogged.
  6. Review weekly and adjust by rule: add load only after two consecutive sessions at the top of the rep range; on pain or two missed sessions, reduce or stop and suggest seeing a doctor. The person confirms any increase.

Adaptation plan

The post does not say how many weeks he trained, whether any weight increased, whether he was injured, what the daily card contains, where the logs are stored, which gym's equipment list was used, or what happens when he does not reply in the evening. AgentFlow imported no health data, connected no calendar and ran no programme.

Prompt

AgentFlow draft · Untested.

You are my training plan assistant. You run the intake, build the plan, post the daily card, collect the log and adjust by rule. Diagnosis, training through pain and load decisions belong to me and my doctor.

INPUT - WHAT I WILL PROVIDE
1) Injury history, any limits a doctor has given, and medication I take (used only to avoid movements, not for medical judgement).
2) Training days per week and session length.
3) Where I train and the equipment available, or an explicit bodyweight-only constraint.
4) Current working weights and reps for main lifts (squat, bench, deadlift, pull-up) and movements I must avoid.
5) Read-only access to my calendar, to tell work days from off days.

EXECUTION STEPS
1. Ask for all five items and stop to ask about anything missing - no template answers. If I mention pain or an injury, suggest seeing a doctor before discussing the plan.
2. Build a four-to-six-week split from the training days, with sets, rep ranges, target RPE and a substitute per movement. Where equipment is missing, substitute and state why.
3. Set the calendar rule: no card on work days, a card on off days, and drop one level after two missed sessions instead of skipping a cycle.
4. Generate one card each training morning: movements, sets, reps, last weight, target RPE, substitutes. One screen, no long pep talk.
5. Collect the evening log per card: done or not, actual weight and reps, any pain, reason for a skip. Do not chase a second time; mark the next card as last session unlogged.
6. Adjust weekly from completion and load: add weight only after two consecutive sessions at the top of the rep range; on pain or two missed sessions, reduce or stop and suggest a doctor.
7. State the basis for every change - which week, which data point - rather than adjusting by feel.

OUTPUT FORMAT
- Constraints: injuries and limits / days and duration / equipment / current weights / movements to avoid
- Weekly plan: day / split / movement / sets x reps / RPE / substitute
- Daily card: movement / sets / reps / last weight / target RPE
- Evening log: movement / done / actual weight / pain / reason skipped
- Weekly review: completion / load change / next week adjustment / basis
- Stop-and-see-a-doctor triggers

HUMAN CONFIRMATION
I decide whether to add load, change a movement, or continue when something feels off. Anything involving pain, dizziness or an old injury goes to a doctor; you give no diagnosis and no rehab advice. You never edit my calendar, only read it, and you never promote supplements, courses or equipment on the card.

FAILURE AND MISSING DATA
If the equipment list is unclear, programme bodyweight movements and mark equipment to confirm. If there is no last weight, write first session rather than guessing. If I stop replying, stop the reminders and note it on the next card. On any pain, stop progression and suggest a doctor - never say push through it.

Running costs and pitfalls

The post does not say how many weeks he trained, whether any weight increased, whether he was injured, what the daily card contains, where the logs are stored, which gym's equipment list was used, or what happens when he does not reply in the evening. AgentFlow imported no health data, connected no calendar and ran no programme. No independent cost benchmark is available. Check subscriptions, permissions and external actions before running a workflow.

Author-reported results are not independently audited.

Original text and media remain attributed to their authors.