How Coach AI Builds Your Weekly Plan: The Case for Every Design Decision

Most AI workout generators work the same way: you pick a goal, a language model writes a week of exercises, and the weights are whatever the model thought sounded right. That approach produces plans that look reasonable and can't be checked. You don't know why it chose 185 instead of 205, and next week it might choose differently for no reason.

Coach AI was built the other way around. The decisions that matter most for results (how heavy, how many sets, when to push, when to back off) come from rules you can write down and defend. The AI handles what it's good at: picking exercises that fit your split and equipment, working around injuries you've told it about, and explaining each prescription in plain language. This post goes through those rules goal by goal and explains the reasoning behind each one.

The Short Version: One Engine, Four Goals

Every number below comes straight from the plan builder. Percentages are of your estimated one-rep max (e1RM), calculated from your logged sets.

Weeks 1–3 are accumulation; week 5 is a deload at about 60% of your recent top set (55–60% for Lose Weight).

One design choice matters here: these bands are stored in one place in the code, and both the AI's instructions and the code that enforces weights read from it. Earlier versions of most AI planners (ours included) had the prompt saying one thing while the validation logic assumed another, and the model would resolve the conflict unpredictably. A single source of truth means the model and the code can't disagree.

Your Log Sets the Weights, Not the Model

This is the part that only works because you log. When Coach AI builds your week, the app sends a summary for every lift you train: your best set from the last 90 days, the heaviest weight you've actually moved in that window, the date, and how many sessions you've logged on it. From the best set, the server calculates your e1RM using the Epley formula (weight × (1 + reps ÷ 30)).

Then it converts your e1RM into this week's working weight using two inputs, the target reps and the target effort, expressed as reps in reserve (RIR), meaning how many more reps you could have done. The percentage is roughly 1 ÷ (1 + (reps + RIR) ÷ 30), clamped to your goal's band and rounded to the nearest 2.5 lb. Here's what that looks like for a lifter whose best bench in the last 90 days is 225 × 5 (e1RM ≈ 262 lb) on the Get Strong goal, intermediate tier:

The AI never decides those numbers. It writes the plan, and then the server overwrites the weights on any lift with a trustworthy e1RM, keeping the model's set structure (top set, working sets, back-offs) but anchoring it to the computed load.

"Trustworthy" is defined strictly. A lift only gets computed loads when it has a set of 1–10 reps in the last 90 days (Epley gets unreliable above 10), at least two logged sessions (one session could be a typo or a bad day), an e1RM that isn't wildly above your heaviest real set, and a ratio to bodyweight that's physically plausible. A lift that fails any of these checks falls back to the model's suggestion, which is still bound by the safety caps covered below. Names have to match exactly, so your dumbbell bench never borrows your barbell bench numbers.

The order of authority is deliberate: your log beats the model, and the safety caps beat both.

Why a 5-Week Block Instead of "Add 5 lb Every Week"

Adding weight every session works for a few months. After that, fatigue builds faster than fitness and the linear approach stalls. The standard fix in strength coaching is periodization: organize training in blocks where you build up on purpose, then reduce load so the fitness you built can show. Meta-analyses comparing periodized and non-periodized programs with matched volume generally find a modest but consistent strength advantage for periodized training (Williams et al., Sports Medicine, 2017).

Coach AI's block is three accumulation weeks, one peak week, and one deload:

1. 1

   Weeks 1–3: Accumulation, with a real ramp

   Effort climbs each week. The intermediate tier goes RPE 7 → 8 → 8.5, and week 3 adds a set to the main lift. If the three weeks were identical, the block would just be the same week repeated, with fatigue building and nothing new to adapt to.

2. 2

   Week 4: A goal-specific peak

   This week is used to realize what you built, and "realize" means something different for each goal (covered below). This is the main difference between the four goals.

3. 3

   Week 5: Deload

   Two sets at about 60% of your recent top weight, RPE around 6. An international expert consensus on deloading (Bell et al., Sports Medicine – Open, 2022) describes the deload as a planned reduction in training demand to manage fatigue and set up the next block. That's how Coach AI uses it.

One implementation detail removes a whole class of bad first impressions: your block starts the week you generate your first plan. An earlier version tied the cycle to the calendar, so a brand-new user could land on "Deload, Week 5/5" on day one. Now each lifter's cycle is anchored to their own start.

Get Strong: Heavy, Specific, and Built Toward One Rep

The decision: 3–5 reps at 82–93% of e1RM. Reps drop 5 → 4 → 3 across accumulation while load rises. Week 4 is a taper with two opener triples at about 88–92% of the attempt weight, then one labeled "PR Attempt" single. Load is the progress metric.

Why it's right: Strength is specific to load. The meta-analysis evidence is fairly consistent that heavier loads produce larger 1RM gains than lighter loads, even when muscle growth is similar (Schoenfeld et al., Journal of Strength and Conditioning Research, 2017). If your goal is a bigger squat, bench, or deadlift, you have to practice moving heavy weight. Dropping reps across the block is standard practice: each week is closer to the single you're aiming for.

How the PR attempt is chosen: This is where log history matters most. The attempt is calculated as about 96% of your e1RM, which should be a single you can make, then limited to no more than 5% above the heaviest weight you've actually moved in the last 90 days. It must also be strictly greater than that heaviest weight. That rules out two failure modes: the model calling a submaximal set a PR, and an optimistic e1RM sending you to a weight you've never been close to. In the example above, the e1RM math alone would suggest 250. The plan prescribes 235 because the heaviest you've moved recently is 225.

Accessories: For intermediate and advanced lifters, accessories get 3–4 sets during accumulation, but never more sets than the day's main lift, and no drop sets or rest-pause. Accumulation is meant to add volume, not extra fatigue per set. Accessories drop to 2 sets in peak and deload weeks. An earlier version capped Get Strong accessories at 2–3 sets all year, which was effectively taper logic running during the weeks meant for building volume. We fixed it.

Build Muscle: Volume Is the Driver, So Volume Is What Peaks

The decision: 8–12 reps at 67–77% of e1RM, 1–3 reps in reserve. Progression uses double progression: add reps until every working set reaches the top of the range, and only then add weight. Week 4 is a volume push with one extra working set per exercise and explicitly no heavy singles. Advanced lifters get 4–5 sets on every isolation exercise and target 16–20 hard sets per muscle per week.

Why it's right: Within a wide range of loads, muscle growth tracks the number of hard sets more closely than the weight on the bar. Dose-response meta-analysis shows more weekly sets per muscle tends to produce more growth, up to a point (Schoenfeld, Ogborn & Krieger, Journal of Sports Sciences, 2017). So the peak week for this goal adds volume. A PR single in a hypertrophy block adds joint stress and fatigue without adding much growth stimulus.

Double progression is the right overload model for the same reason. A set of 12 at 185 is more stimulus than a set of 10 at 185, and getting from 10 to 12 reps is usually a more achievable step than jumping to 190. Adding weight every 1–3 weeks rather than every session keeps rep quality high, and Coach AI flags dropped reps as the problem to fix, not a flat top weight.

Why volume is stacked on accessories: For advanced lifters, extra volume goes on isolation work, not more heavy squats and deadlifts. Isolation sets build the target muscle at a fraction of the whole-body fatigue a heavy compound set costs, so you can add volume and still recover.

Lose Weight: The Job Is to Keep Your Muscle

The decision: 12–15 reps at 60–70% of e1RM, effort capped at RPE 8, three sets on the main lift. Accumulation weeks are labeled maintenance and hold current loads, so success means completing every set with good form, not adding weight. Week-over-week increases are capped at 2.5%, half the cap the other goals get. Plans make sessions harder with short rest (45–60 seconds) and circuit-style ordering, not more sets or heavier weight. Deloads go to 55–60%.

Why it's right: In a calorie deficit, recovery is the limiting factor. Research on energy deficits and resistance training finds that a deficit significantly blunts lean-mass gains, while strength is far less affected (Murphy & Koehler, Scandinavian Journal of Medicine & Science in Sports, 2022). That defines the goal clearly: lifting during a cut is there to tell your body to keep the muscle it has. Holding heavy-enough loads and completing your sets does that. Chasing PRs on reduced food adds fatigue you can't recover from, and a lifter who fails sets every week loses confidence in the plan.

That's also why Coach AI analysis reads this goal differently. For a Lose Weight lifter, an unchanged top weight in a maintenance week counts as a success. What gets flagged is declining completion, and a strength drop of more than 5% over two weeks, which can mean the deficit is too aggressive. Short rest and density increase the training demand without adding to the muscle-damage and recovery cost that more volume would.

General Fitness: Built for Showing Up Every Week

The decision: 8–12 reps at 65–75% of e1RM, three sets, effort capped at RPE 8. Progression is optional: add one rep or 2.5–5 lb only if the last session felt easy, never forced, at most every 2–4 weeks. There is no peak week; week 4 is another steady week. Week 5 is a light week.

Why it's right: The biggest predictor of results for someone training for general health is whether they're still training in six months. A program that tries to max you out treats a general-fitness lifter like a competitor, and that lifter will usually stop showing up. Moderate reps across a mix of movement patterns cover strength, muscle, and joint health without making every session feel like a test. Leaving out a peak week was deliberate: this goal has no performance to peak for, so a fifth heavy week would only add stress.

Rules That Apply to Every Goal

Some decisions don't depend on the goal. They protect every plan from the most common ways auto-generated programs hurt people or waste their time.

Your experience level is measured, not guessed

Coach AI places you in a tier (easy, moderate, high) mainly by comparing your e1RMs on the core barbell lifts (squat, bench, deadlift, overhead press, row) to strength-to-bodyweight standards, using your best three. Machine and isolation lifts are left out because their generous standards make beginners look advanced. Training age can raise the floor (3+ years means at least moderate) or lower the ceiling (under a year caps you at moderate). Total session count is used only if there's no usable strength data. Tier sets the effort ramp and set ranges, and you can always override it.

Weight jumps are capped by code

A top set can rise only about 2.5–5% over your heaviest recent weight for that lift (2.5% for Lose Weight). Behind that, there's a hard single-session limit (10% for Get Strong) that no prescription can exceed for any reason. There's also a lower bound: if a computed set comes out below 85% of the engine's weight because of a rounding error or bad math, it gets raised back. Plans can be wrong in either direction, and the code checks both.

Weekly volume has a ceiling

Hard sets are counted per muscle across the week. The first muscle listed for an exercise counts as a full set and any others count as a half set. The ceiling is 20 hard sets per muscle per week. Above that, the server trims accessory sets, never main lifts. If you've flagged a muscle as lagging, it gets priority placement, at least two sessions that week, and a ceiling of 26 sets. That's the one place we allow more volume, because you asked for it.

Missed sessions mean shorter workouts, not lighter ones

If last week's adherence was low, the plan keeps your intensity and cuts session length: it drops the lowest-priority accessory, aims for 60 minutes or less, and keeps every main-lift set. Missing workouts is usually a schedule problem, not a recovery problem. Lowering your loads because you were busy would cost you strength you didn't need to lose. An earlier version did exactly that, and we changed it.

Injuries and preferences are binding

If you've told Coach AI about an injury or a movement to avoid, that movement doesn't appear anywhere in the plan. It gets substituted with something that trains the same muscle, and the swap is explained. If you've said "hold my squat at 225," the plan treats that as a rule.

Why This Works Better the Longer You Log

Every decision above depends on one input: your training history. That's the main reason Coach AI produces better plans than a template, and why the plan improves the longer you use KOBP.

• More of your plan is computed. On day one, with fewer than three logged sessions, you get a conservative starter plan. Once a lift has two logged sessions and a recent set in the 1–10 rep range, its loads come from your own numbers. Over time, nearly every lift you train is computed from your data.
• Exercises stay the same so progress is measurable. In block weeks 2–5 the plan keeps last week's exercises in the same day order. A progression is only meaningful if you're comparing the same lift week to week. At the start of each new block, main lifts stay and about half the accessories rotate to new variations for the same muscles, enough variety to avoid boredom and overuse without losing the thread.
• Your reported effort adjusts the next week. If the RPE you've logged over your recent sessions averages 9.5 or higher, next week's effort target drops a full point. If it averages 6.5 or lower, the target goes up half a point. The plan responds to how training actually felt.
• Your analysis knows your plan. Each logged session is tagged with its block phase, so Coach AI's analysis doesn't treat a planned deload as a decline or a Build Muscle volume week as a stall. The text that tells the plan builder what a phase means is the same text that tells the analyzer how to judge it. (We covered why that matters in Why Most AI Coaches Misread Your PRs and Deloads.)
• Each plan's results are recorded. Every plan is stored with its phase, tier, and settings, then compared against what happened: what share of sessions you completed and how your e1RM moved the next week. That's how we check whether these rules work for real lifters, not just on paper.

A template gives everyone the same plan. A chatbot gives you a plan based on whatever you typed that day. A plan built from a complete, consistent log starts from what you actually lifted, and every week of logging gives it more to work with.

What It Doesn't Do

Being clear about the limits is part of the case.

• It doesn't adjust set by set during a workout. Targets are set per week; your logged RPE shapes next week, not the next set.
• It doesn't estimate maxes from high-rep sets. Anything above 10 reps is excluded from e1RM, so a lift you only train for 15s gets the model's suggestion plus safety caps, not computed loads.
• It doesn't replace a coach watching your form. It can ask you to prioritize form; it can't see it.
• It doesn't manage your nutrition. Lose Weight programming protects muscle during a deficit; it doesn't create the deficit.

A Plan Built From Your Own Log

Log free with KOBP Strength. When you're ready, Coach AI turns that history into a 5-week block built for your goal, with weights computed from your actual lifts.

5-Week Blocks e1RM-Based Loads Goal-Specific Peaks Jump Caps Injury-Aware