The question trainers most want answered about AI programming tools isn't philosophical — it's practical. Is the output good enough to use? Does it look like something I'd actually write? The best way to answer that is to put the two approaches side by side for the same client and see what each produces.
The client
The client for this comparison is a 36-year-old man, four years of consistent resistance training, training four days per week in a fully equipped commercial gym. Current goals: increase squat and deadlift strength, maintain body composition. Relevant history: mild lower back sensitivity managed successfully through limiting deadlift volume to three to four working sets per session. Current benchmarks: squat at 110kg for five reps at RPE eight, deadlift at 130kg for three reps at RPE eight. Moving into week one of a new six-week block.
The manually written session: Day 1 (lower body)
The trainer writes this session from scratch, drawing on four years of programming history with this client and their own accumulated methodology. Primary movement: barbell back squat, four sets of four at RPE seven — conservative in week one, building load tolerance before increasing intensity in week three. Secondary movement: Romanian deadlift, three sets of eight at moderate load — hip hinge volume that doesn't load the lower back as heavily as the conventional pull. Accessories: leg press, three sets of ten; walking lunges, two sets of twelve each side; leg curl, three sets of twelve; ab wheel rollout, three sets of eight. Total working sets: fifteen. Time to write: approximately forty-five minutes, including thinking through the block structure for weeks one through four and cross-referencing previous block notes.
The AI-generated session: Day 1 (lower body)
The AI generates this session from the client profile, session history, and trainer's programming preferences. Primary movement: barbell back squat, four sets of four at RPE seven. Secondary movement: Romanian deadlift, three sets of eight at moderate load. These two entries are identical to the manually written session. Accessories: leg press, three sets of ten; Bulgarian split squat, two sets of ten each side; lying leg curl, three sets of twelve; cable pull-through, three sets of twelve. Total working sets: fifteen.
Time to generate and review: approximately eight minutes. The trainer reviews the primary and secondary movements (confirmed), checks the loading scheme (appropriate), and makes one targeted swap — replacing the cable pull-through with an ab wheel rollout because this client has historically responded better to anti-extension core work than hip hinge posterior chain accessories at this point in a block. That reasoning comes from session notes that aren't yet in the AI's context window. The review and edit takes four minutes.
What the comparison reveals
The structure and the high-value programming decisions are identical. Four sets of four on the squat at RPE seven in week one of a strength block for a client with this profile is the right call regardless of how it's generated. The Romanian deadlift selection as the secondary movement reflects the lower back management approach that both the trainer and the AI apply correctly. The accessories differ in one entry — a reasonable alternative that the trainer overrides based on client-specific session history that wasn't in the system.
The time difference is the most significant variable: forty-five minutes versus twelve. That gap does not reflect a difference in programming quality for this client — the delivered session is essentially identical. It reflects the difference between construction and review as the primary task.
Where the comparison is most favorable to the AI
The comparison above represents a well-profiled client with substantial session history in the system. For clients at this stage, with this level of data available, the AI output is close enough to the trainer's own work that the editing requirement is minimal and the time savings are substantial.
Where the comparison is most favorable to the manual approach
For a new client with limited session history and no established profile, the AI output requires more revision. The trainer's intuition about this person — built from assessment sessions, initial conversations, and the first few weeks of coaching — isn't fully captured in the data yet. The manual approach draws on that unstructured knowledge. The AI approaches the same session with more generic priors. The gap narrows as context accumulates, but in the first few weeks of a new client relationship, the manual approach may produce a first draft that requires less revision than the AI output.