The Supplemental AI Philosophy

Quickstart

Most fitness apps that talk about AI mean a tool that decides for you. It picks your next workout. It tells you to rest. It sets the weight for tomorrow’s bench. Kinoku does the opposite. The AI here shows you patterns. Like this one: “Your bench top sets have gotten harder at the same weight for four weeks.” It does not decide what to do next. You do.

This is not just a design choice. It is a stance on what AI is good at and where it falls down. Letting software run your training tends to make the training worse, even when the software reads the data right.

Two kinds of AI in fitness

It helps to keep these two apart.

Replacement AI

The app runs your training. It picks your next workout. It changes sets and reps based on “readiness.” It tells you when to back off. It swaps exercises on its own. Fitbod, Future, and the auto-coaches in most apps work this way. The pitch is simple: “You don’t have to think about it.”

Supplemental AI

The app shows you patterns you would miss. “Your chest volume is 40% below your weekly target this week.” “Your Pulse Score is low because your recovery dipped sharply yesterday.” “Your effort on top-set bench has crept up for four weeks at the same weight.” It does not act on any of this. It just makes sure you see it.

Both have a place. But they are not the same. For people who train themselves, supplemental AI is almost always the better default.

Why replacement AI struggles with strength training

Replacement AI works well for endurance sports like cycling or running with a power meter. It works there because the effort is measured cleanly by hardware. The body’s response is well studied. And the goal is usually a race time the model can aim for. Tools like TrainerRoad’s adaptive plans and Garmin’s daily suggested workouts do a decent job because the problem fits.

Strength training has none of these traits.

The data is a guess

Effort scores are self-reported. “Two reps left in the tank” means whatever you judged it to mean today. That judgment shifts with sleep, the lift, the time of day, even your coffee. A model trained on those numbers is trained on noise with a faint signal inside.

People differ too much

Lifters respond to the same training in wildly different ways. A program that works for one person stalls another at the same level. AI tries to learn your own response curve. But the data it would need is usually more than your whole training history holds.

Context is invisible

You had a fight with your partner. You traveled yesterday. You ate badly. You feel a cold coming on. The app only sees “3 sets of 5 at 180 kg, hard effort.” The real story of that session does not fit in a data field. And that story is often what matters most for tomorrow.

The goal keeps moving

A runner has a race. A lifter might want a meet, a PR, a look, or all three at once. And that mix shifts every few months. An AI aiming for “strength gains” may be chasing a goal you will not even want in six weeks.

Add these up and replacement AI has a poor fit. It can give an answer that looks fine but has no idea when it is wrong.

What supplemental AI does well

Here is the key idea. AI is great at the hard part for humans: spotting patterns across long stretches of messy data. It is weak at the human part: judging that pattern against your real life.

A lifter usually cannot catch these on their own:

A muscle group’s weekly volume has drifted 30% below target over the last month
Top-set bench effort has crept up at the same weight for 4 weeks
Short-term fatigue has stayed high for 21 days with no drop
The bench-to-squat ratio has been out of range for 8 weeks
Run pace has flatlined while distance kept rising

These hide in thousands of data points. An AI that shows them each morning is a real help. An AI that acts on them is not. Cutting your weight, swapping a lift, or canceling your session on its own is often wrong. It cannot tell if you just got back from vacation, or whether that effort climb is real fatigue or just a shift in how you score it.

The rule is simple. The app detects and shows. You decide. The detection runs on its own. The choice stays yours.

How Kinoku builds this

Every AI-driven part of Kinoku follows the same rule.

Pulse Score

A single score from 0 to 100 that blends your training, wellness, and health signals. Pulse shows your state for the day. It does not tell you what to do. On PRO it shows three insights per day, pulled from a library of pattern types. You read each one. You decide.

Pulse also handles missing data well. If you skip wellness logging, it shifts the weight to what you do have. It does not fake confidence when the data is thin.

Coach screen

Today’s Focus and Weekly Recap come free. Active Alerts and Exercise Intel come with PRO. The Patterns engine comes with ELITE. Every card has a Why? panel. It shows the inputs, the source, and how sure the app is. You can disagree. The app does not mind.

Forecast insights

These look 1 to 3 days ahead for things like overreach, rest debt, and cycle-phase shifts. They read like this: “Your form trend points to an overreach in about 2 days if your load holds.” Not “rest tomorrow.”

Readiness Adaptation (ELITE, optional)

This is the closest Kinoku comes to replacement AI, and you have to turn it on yourself. When it is on, Readiness can pre-adjust tomorrow’s weights and reps. It uses your recovery and recent load to do so. Even then, the change is a suggestion. The app shows the proposed change and the reasoning. You approve or edit it before the session.

There is a hard limit. Readiness can shrink a session by up to 15%, but it can never cancel it for you. You decide whether to rest. The app only flags that it might be a good day for it.

Autoregulation

After each set, Kinoku can tweak the next set based on how many reps you had left. This is the most active of these features. It works set by set, and you stay in the loop the whole time. You log the effort, the app suggests the next load. You can ignore it. The app still works.

The pitch, in plain terms

The usual fitness pitch is: “An AI coach. Plans that adapt to you on their own.” Kinoku’s pitch is: “The AI shows you what you can’t see. You still call the shots.”

It is a less exciting pitch. It promises no magic. But it leads to better training for the kind of person Kinoku is for. That person has lifted for years, has views on their own program, and wants a tool that sharpens what they notice. Not one that takes over their judgment.

Flip it around and the stakes get clear. Every time a fitness AI “adapts your plan,” it takes a stand on what you should do next. Those calls run on thin information. Stacked up over a year, they drift toward the middle: moderate volume, moderate effort, moderate everything. That is fine for the average user. It quietly hurts anyone chasing something specific.

What this means for the app

A few results follow from this.

Fewer push alerts. Kinoku does not ping you to rest. It shows you in the app when you open it.
No “Daily Suggested Workout” on the home screen. Smart Today shows what you can do, based on your rotation and recovery. It does not push.
Everything gets a reason. Every AI-driven card has a Why? panel. If you can’t see the inputs, you can’t really agree or disagree.
You can turn any AI part off. Hide Coach. Keep Pulse minimal. Ignore Forecast. If you just want to log workouts, Kinoku works as a plain log.

Here is the quiet part most apps won’t say. The best fitness AI right now stays out of your way most of the time. It speaks up only when it spots something you would have missed. Kinoku tries to be that.

Track this in Kinoku

Most of what’s here is free: base Pulse Score, Coach Today’s Focus and Weekly Recap, and a Why? panel on every insight. The deeper layers live in Elite Analytics and the Patterns tier of AI Coach. The stance stays the same across every tier.