The fitness industry is full of "AI-powered" programs that are little more than dressed-up spreadsheets. A truly algorithmic training program does something fundamentally different: it reads your data, applies peer-reviewed research, and generates plans that adapt to your physiology in real time. Here's how it actually works under the hood.
MUSCLE TECHNICS uses a Large Language Model (LLM) constrained by fixed scientific rules. Unlike a simple if-then algorithm, the LLM receives your complete training context (workout history, recovery status, plateaus, volume trends) and generates a plan while strictly following evidence-based guidelines.
The meta-analysis of 67 studies defines optimal set counts by experience level: beginners 6-10, intermediates 10-16, advanced 16-22 sets per muscle group per week. The AI counts every set toward all the muscles involved, secondary muscles partially.
Training each muscle group twice per week produces significantly more hypertrophy than once. For the split you choose, the app recommends the number of training days that hits every large muscle group at least twice a week.
A meta-analysis of 15 studies (Refalo et al. 2023) shows that RIR 1-3 produces practically the same hypertrophy as training to failure, with less fatigue (Refalo et al. 2024). The AI prescribes RIR targets per set: beginners RIR 2–3 on every set; intermediate and advanced lifters RIR 2 on the first set, RIR 1–2 on middle sets and RIR 1 on the final set, one rep short of failure, isolation exercises included. From age 60, it plans one more rep in reserve: RIR 2 on every set, RIR 3 for beginners.
Your exercises stay the same from session to session, so your progress stays measurable. An exercise is only swapped once it stops setting new bests and an attempt to increase brings nothing either, and then for one with the same movement pattern that works the same main muscle.
Each muscle has its own recovery timeline, modified by individual factors. The AI computes exact hours since last training for each muscle group, compares against the required recovery window (adjusted for age), gives muscles that aren't fully recovered fewer sets in today's plan and leaves out those that clearly aren't.
When you tap "Create Plan," here's the sequence:
Data collection (~200ms): The app queries your last 20 workouts with all sets, personal records, weekly volume, 4-week trends, the exercises you've excluded, and recovery timestamps per muscle group.
Recovery classification (~50ms): Each muscle is marked as "recovered", "almost recovered" (fewer sets) or "excluded" based on hours since last training vs. required recovery, with age modifier.
AI generation (~3 seconds): All data plus the rules are sent to the LLM. For the day that's due in your chosen split, it chooses the exercises from scientifically rated options; sets, rep ranges (per Schoenfeld 2021), and RIR targets are set by fixed rules. It also generates a greeting with context, a week plan and an explanation of the plan.
Validation: The response is checked for duplicate exercises, valid exercise IDs, and proper JSON structure. If anything fails, a deterministic fallback plan is generated using pure algorithmic logic.
The AI tracks your consecutive training weeks and adjusts volume from what it measures:
A muscle group keeps falling: if your e1RM in a muscle group clearly drops, the AI removes sets from that group, while every exercise keeps a minimum number of sets.
Several weeks in a row plus high fatigue: if your recovery check-in reports high fatigue after you have trained for several weeks in a row, the whole session becomes a deload with fewer sets, while every exercise keeps a minimum number of sets and the starting weight is not raised.
Moderate fatigue: the session keeps its structure with slightly fewer sets.
Otherwise: the coach picks sets within the normal range for your level. It is told how many weeks in a row you have trained: early on it may add volume, after a longer stretch it is told to hold. There is no fixed cycle length and no scheduled deload week.
The app tracks your e1RM (estimated one-rep max) for every exercise. If an exercise stops setting new bests, it flags stagnation and responds in two steps, starting with an attempt to increase:
First an attempt, then a swap: The app first tries an increase, clearly marked in your plan. Only if that brings no new best does it swap the exercise for one with the same movement pattern and the same main muscle. Barbell bench still stuck? The dumbbell bench press takes its place. If your performance is falling instead, the exercise stays and the plan reminds you of sleep and recovery.
No rep-range switching: The rep range for each exercise is set by fixed rules and stays the same, so your e1RM trend stays comparable. The AI does not switch rep ranges as a plateau response.
Muscle-specific deload: Instead of deloading everything, the AI can reduce volume for just the muscle group whose performance is falling while maintaining progress elsewhere.
A spreadsheet follows a fixed path regardless of your performance. An AI program reads your actual data (every set, every rep, every RIR) and generates a new path each day. It's the difference between a bus route and a GPS.
No. The AI handles all the complexity. You just log your sets and the coach does the rest. But every recommendation includes an explanation, so you learn over time.
Every plan has a deterministic fallback. If the AI output fails validation, a rule-based algorithm generates the plan instead. And the scientific rules constrain the AI: It cannot recommend anything that violates the research.
→ Evidence-Based Training Guide
Fixed rules, peer-reviewed studies, Full exercise library. Your training data in, science-based plan out. Every day.
Try free for 14 days →