In 2026, generating a training programme with artificial intelligence takes under thirty seconds. Describe a goal, weekly frequency and experience level, and out comes a formatted plan with sets and reps, carrying an appearance of competence that fools a lot of people.
The tool is not the problem. What people conclude from it is.
What AI genuinely solves
I use artificial intelligence daily, and it has changed my operation concretely. Not in the way the advertising promises.
Periodisation modelling and verification
When designing a training cycle, a set of relationships has to add up: weekly volume per muscle group, intensity distribution, session density and recovery time between stimuli.
Checking that by hand across dozens of clients consumes hours. AI verifies the internal coherence of a design in seconds and flags inconsistencies: a muscle group with disproportionate volume, insufficient recovery for the prescribed intensity, a progression that accelerates too early.
It flags. The professional still decides the final design.
Reading historical data series
For me, this is the most valuable use.
A client coached over two years generates a great deal of information: skinfolds, girth measurements, loads, tests, training frequency, reports on sleep and energy. Looking at one assessment tells you little. Looking at twelve assessments together reveals patterns: the month progress stalled, the correlation between dropping attendance and stagnation, the exact point where load progression stopped tracking strength gains.
Those patterns were in the data all along. The difference is that they are now visible.
Organisation and reclaimed time
Records, protocols, programme sheets, reports. Necessary work that does not require human presence.
Every hour technology recovers here is an hour returned to what does require presence: watching a client execute, correcting, talking, understanding why they missed three sessions last month.
What AI does not do, and will not do soon
This is where the conversation gets uncomfortable for anyone selling "AI-generated training".
AI does not assess pain. It cannot distinguish the normal discomfort of a new stimulus from pain signalling injury. That reading depends on asking, watching the client's face, testing range and comparing against their history.
AI does not read posture in real time. Computer vision tools are useful for movement analysis, and I use them as support, particularly in online coaching. But identifying a subtle compensation mid-set, with the client under load, is a different problem.
AI does not perceive human context. The client who slept four hours, who is grieving, who is afraid to lift again after an injury. None of that appears in a prompt, and all of it changes the session.
AI does not carry technical accountability. This is the central point, and the least discussed.
The accountability question
In Brazil, exercise prescription is a restricted activity of Physical Education professionals registered with the regional council. That is not bureaucratic formality: it assigns technical responsibility to an identifiable person who answers for what they prescribe.
When someone hands over an automatically generated programme, with no prior assessment and no review, the question is simple: if the client is injured following that plan, who is accountable?
An algorithm holds no professional registration. It has no duty of care. It cannot be questioned about why it prescribed a given load to a given person.
The model that makes sense
The configuration I argue for is straightforward:
- Assessment is human. Always, and first.
- Analysis may be assisted. AI processes volume and surfaces patterns.
- The decision is human. The professional interprets, validates or discards.
- The deliverable is signed. Every programme is reviewed and has a named, registered professional behind it.
- Coaching is human. Correction, adjustment and contextual reading happen in the relationship, not in software.
In that arrangement, technology does not replace the professional, it extends their reach. It is the difference between a physician using imaging and imaging replacing the physician.
On data
One final point, and not a small one.
Health data is sensitive data under Brazilian data protection law, as it is under the GDPR. Entering identifiable client information into third-party AI tools, without a legal basis and without proper consent, is a genuine legal problem, and an ethical one before that.
Anyone using AI in this field needs to know exactly what information they are sending, where, and under what terms.
Conclusion
Artificial intelligence is the most useful instrument to reach exercise science in a long time. It is also the easiest to misuse.
The right question is not "do you use AI?". It is: where does it start, where does it stop, and who signs the result.