Blog · Programming · 8 min read · Updated August 2026

Letting an AI Plan Your Training Block

Give a modern AI your real training history and ask for a four-week block and you'll get something that looks like a coach wrote it. The interesting question isn't whether it's plausible — it's where the plausibility stops and the judgment should start.

What it reliably gets right

Structure

Splits, weekly layout, session ordering, and the shape of a progression are well-trodden ground, and AI handles them well. Ask for four days built around the squat and you'll get a coherent week: heavy squat early, an upper day after, a lighter squat variation mid-week, accessories placed where they won't interfere. This is the part that takes a novice hours and an AI seconds, and the output is genuinely fine.

Spotting what you've been avoiding

This is where a data connection earns its keep and where AI often beats self-assessment. Everyone drifts toward what they like. Given three months of actual sessions, an AI will notice — with receipts — that you've done twice as much pressing as pulling, that your posterior chain work quietly disappeared six weeks ago, or that "leg day" has been the one you skip when the week gets tight. It doesn't flatter you, because it has no reason to.

Reworking around constraints

Travel weeks, a closed gym, a week with three days instead of five — rebuilding a plan around a constraint is tedious bookkeeping, and that's precisely what this is good at. Tell it you have dumbbells up to 50 lb and a pull-up bar and it will produce something reasonable in one pass, where doing it by hand means twenty minutes of substitution lookups.

Explaining itself

Underrated: you can ask why. Why this rep range, why front squats here, why the volume drops in week three. The explanations are generally sound and consistent with the plan, which makes the plan easier to evaluate — and easier to learn from.

Where it consistently falls down

Load selection near your limits

Percentages of a max are arithmetic and AI does them fine. Judging whether you can hit a hard triple at 90% on a Thursday after a bad night's sleep is not arithmetic. AI tends toward the textbook answer, which is usually slightly optimistic for the back half of a block, when accumulated fatigue is real but invisible in the numbers it can see. Expect to shave the top end.

Injuries and anything it can't see

An AI honors the constraints you state. It cannot infer the ones you don't. If your left shoulder complains on overhead pressing and you've never written that down anywhere, you'll get a plan with a lot of overhead pressing — and it will look well-reasoned, because from the model's view it is. Say the quiet constraints out loud, every time.

Over-engineering

Asked for a program, AI tends to produce a program — undulating percentages, four accessory variations, a specific tempo prescription. Complexity signals effort, and models are rewarded for looking thorough. But most people's progress isn't limited by their programming's sophistication; it's limited by turning up and adding weight. If the plan looks impressive, that's often a reason to simplify it, not to be reassured.

Knowing when to abandon the plan

A block is a hypothesis. Two weeks in, reality has an opinion. The instinct to tear up week three because you're clearly cooked — or to push harder because it's moving faster than expected — is the coaching judgment that AI is furthest from. It will happily help you adapt the plan once you decide to. Deciding is still yours.

How to prompt around the failure modes

Most of the bad output traces back to missing context or an unexamined assumption. Four habits that fix most of it:

• Make it read before it writes. "Look at my last 8 weeks first, then tell me what you see, then propose the block." Forcing the read as a separate step produces a visibly better plan than asking for the plan cold — and lets you catch a bad read before it becomes a bad program.
• State every constraint, including the embarrassing ones. Days available, equipment, injuries, the fact that you will not do 45 minutes of accessories. It plans for the athlete you describe, so describe the real one.
• Ask it to argue against itself. "What's the weakest part of this plan? What would a skeptical coach say?" This reliably surfaces the over-engineering, and it's the fastest way to find the thing you'd have discovered in week three.
• Cap the complexity explicitly. "Five exercises per session, no more" gets you something you'll actually run.

The honest summary

AI is a fast, well-read training partner with perfect recall of your log and no instinct for your body. That combination is genuinely useful — it removes most of the tedium of programming and catches imbalances you'd rationalize away — and it is not a coach.

The most reliable way to use it: let it produce the first draft from your real data, interrogate the draft, cut a third of it, run it, and re-decide every couple of weeks with fresh numbers in front of you. That's a workflow where the AI does what it's good at and you do what it isn't.

None of this works without the data, though. An AI programming from an empty log is just a template generator with extra steps. The version worth using is the one that's actually seen your training — which means either a coach built into the app that holds your data, or connecting your own AI to it.