Knowing about AI is not the same as being fit with it
You can follow every new model release, understand how transformers work, know all the latest prompting techniques, and still not be particularly AI-fit.
The same is true of physical fitness. Reading Starting Strength does not make you strong, and studying running physiology does not give you endurance. Fitness is developed by doing.
AI fitness is your demonstrated ability to use AI to think, learn, create, solve problems, and get things done.
The more AI-fit you become, the larger the space of things you can do. That demonstrated capability is what the AI Fitness Assessment tries to measure.
The AI Fitness Model
Physical fitness is not one number. Strength, endurance, mobility, coordination, and power are different qualities. A distance runner, gymnast, and powerlifter can all be extremely fit while having radically different profiles.
AI fitness works the same way. Someone might be extraordinary at building software with AI but rarely use it for research or creative work. Another person might use AI constantly for learning and thinking but never delegate a task to an agent. Neither profile can be adequately described by a single “AI proficiency” score.
The model distinguishes fundamental capacities from compound qualities that emerge when those capacities work together.
Fundamental capacities
These eight capacities describe the underlying dimensions of AI fitness.
Fluency
Can you work fluidly with AI?
Fluency is the ability to communicate intent, provide useful context, iterate, correct misunderstandings, give examples, break down problems, and steer AI toward the result you actually want.
It is less like memorizing “good prompts” and more like developing coordination. With practice, the interaction stops feeling like issuing commands to a machine and starts becoming a natural way of working.
Strength
How much can you accomplish with AI?
Strength measures the magnitude of useful work you can take on. There is a large difference between asking AI to rewrite an email and using it to analyze a difficult problem, create a substantial work product, research an unfamiliar field, or build software you could not have built yourself.
As with physical strength, you develop it through progressive overload: give AI increasingly substantial real problems and learn how to carry them through.
Range
How many different kinds of problems can you tackle?
Range is the AI equivalent of mobility. You might use AI to learn, research, write, analyze data, work with files, create images, build software, plan, organize, automate tasks, or solve everyday problems.
Someone can be extremely capable within one narrow movement pattern while having little idea what AI can do elsewhere. A specialist may also rationally train a narrow range because that is what their sport requires.
Endurance
How much complexity can you sustain?
Many AI interactions are short: ask a question, get an answer, leave. More difficult work requires maintaining direction through exploration, files, research, revisions, mistakes, intermediate results, debugging, and multiple sessions.
AI endurance is the ability to stay with a complex AI-assisted task until it becomes a finished outcome rather than stopping after the first useful response.
Agency
How much execution can you delegate?
AI use can progress through several stages:
Tell me → Help me → Make it → Do it with me → Do it for me.
At the low end, AI provides information and advice. At higher levels, it creates artifacts, operates tools, modifies environments, executes multi-step tasks, and works toward outcomes under your supervision.
Agency measures how far you have developed along that progression.
Integration
How deeply is AI connected to your real work and life?
A chatbot can remain an isolated destination where you manually paste information and carry the results somewhere else. More integrated AI can work with the files, projects, applications, information, and workflows where the real problem already exists.
Integration measures the distance between “using an AI website” and having AI function as part of your actual working environment.
Judgment
Do you know when to trust, verify, intervene, or stop?
Greater AI capability creates greater leverage, but also greater potential for mistakes.
Judgment means recognizing when an answer needs verification, understanding what context AI does and does not have, giving an agent appropriate permissions, keeping destructive actions reversible, noticing when a model has gone off course, and deciding how much autonomy a particular task deserves.
An advanced user with poor judgment can be like a strong lifter with terrible form: capable of moving a lot of weight, but capable of doing damage too.
AI Sense
Do you notice when AI could help?
This is the capacity that turns AI from a tool you occasionally remember into a new way of seeing problems.
A repetitive task appears and you wonder whether AI could automate it. You encounter a pile of documents and realize AI could analyze them. You wish a small piece of software existed and realize you may be able to build it. Someone describes a problem outside your expertise and you recognize that AI might let you cross that boundary.
As AI Sense develops, these opportunities become increasingly obvious without someone else having to point them out.
Compound capacities
Not every useful athletic quality needs to be a primitive dimension. A sprinter’s explosiveness depends on several physical capacities, but power is still useful to observe, train, and discuss. AI fitness has similar compound qualities.
Power
The ability to move rapidly from an idea or problem to a useful result. It combines strength, fluency, AI Sense, and often agency.
Leverage
The amount of useful output you can produce relative to your own effort. More agents or more expensive tools do not create leverage if you spend all day babysitting them.
Orchestration
The ability to coordinate tools, models, agents, and parallel lines of work toward a larger objective.
Adaptability
The ability to find another route when the first AI approach fails rather than simply giving up.
Versatility
The ability to move comfortably among different kinds of AI problems and working modes.
Independence
The ability to approach novel problems with AI without needing someone else to prescribe the exact tool, prompt, or procedure.
These qualities emerge from combinations of the fundamental capacities, but they are often easier to recognize in real behavior.
Different athletes need different fitness
There is no single ideal fitness profile.
A powerlifter and a gymnast train for different outcomes. An elite athlete may be intentionally unbalanced because performance in one sport matters more than general fitness.
AI is similar. A software engineer may develop extraordinary strength, endurance, agency, and orchestration around coding. A researcher may emphasize fluency, endurance, judgment, and range. A business owner may benefit most from AI Sense, leverage, integration, and agency. A creative professional may develop a completely different profile.
For most people, broad general AI preparedness is useful. For specialists, the right question is whether their profile matches what they are trying to accomplish. A low score in an irrelevant capacity is not necessarily a deficiency.
Personal trainer versus sport coach
A specialist may know far more than an AI fitness trainer about their particular “sport”: AI filmmaking, software development, legal research, sales automation, or another domain.
That does not make general training irrelevant. Specialized expertise can coexist with gaps in transferable capabilities such as range, judgment, integration, and adaptability.
An AI sport coach helps improve performance in one application. An AI fitness trainer assesses and develops the person’s more general capacity to work with AI across applications.
What the assessment measures
The AI Fitness Assessment is deliberately not a test of AI trivia. It asks about what you actually do.
Can you steer AI when the first answer is wrong? Do you use it only for questions or also to create substantial things? Can you sustain a difficult project? Have you delegated execution? Is AI connected to your real workflows? Do you recognize new opportunities for using it? Do you know when supervision matters?
Those behaviors provide evidence about the shape of your current capabilities and where additional training could create the most useful adaptation—not a declaration that you are “72% good at AI.”
The assessment is one evolving implementation of the model described here; the model may continue to develop with it.
Training, not merely using AI
Using ChatGPT on your phone to answer everyday questions is a little like walking. It is useful, it keeps AI present in your life, and it is far better than being sedentary.
But repeating the same easy movements eventually stops producing much adaptation. AI Fitness is not just about catching up: once a behavior becomes easy and habitual, further improvement requires a new challenge.
You develop AI fitness by attempting tasks slightly beyond what you can already do comfortably: trying an unfamiliar kind of problem, carrying a project further, giving AI a larger task, connecting it to real information, or delegating more of an outcome while learning how to supervise it.
That is progressive overload for AI.
AI activity and AI fitness are related, but they are not the same thing.
Equipment is not fitness either
Having the most expensive AI subscription does not make you AI-fit any more than owning a home gym makes you physically fit.
ChatGPT, Codex, agents, image generators, research tools, and connectors are equipment. Appropriate equipment matters, but fitness is demonstrated by what you can actually do with it.
The assessment therefore focuses primarily on behavior and capability rather than AI trivia or knowledge of product names.
From assessment to training
Assessment is only useful if it changes what you do next.
A person with limited Range might deliberately try unfamiliar AI movements.
Someone low in Strength might progressively give AI larger real-world tasks.
Someone low in Endurance might take one project all the way from vague idea to completed result.
Someone low in Agency might progress from asking AI for instructions to delegating bounded tasks.
Someone low in Integration might begin working with real files and workflows rather than isolated chats.
Someone low in Judgment might practice verification, reversibility, permission boundaries, and appropriate supervision.
Someone low in AI Sense might examine everyday frustrations and ask: Could AI help me do this differently?
All of it should be practiced with good form. The objective is not dependence on an AI trainer: training should make you increasingly capable of recognizing and solving your own problems with AI.