AI is a muscle you build, not a class you take.
That is the simplest answer to the question I keep getting. People ask me, over and over again, how do I get better at AI? What class can I take? What course is going to make me competent? What is the cleanest path from feeling behind to feeling fluent?
I understand the question. Most professional skills have been packaged that way for years. You take the class. You watch the module. You get the certificate. You come away with vocabulary, a framework, and maybe a little more confidence than you had before.
But that is not how you get good at AI.
How do you get better at AI?
You get better at AI by using it repeatedly on real work. A class can teach vocabulary and basic concepts, but practical fluency comes from writing prompts, getting imperfect outputs, correcting the tool, adding context, and repeating the loop until judgment improves.
It is almost like asking how to get better at running. I do not know. Run.
That answer sounds almost too simple, but it is the part people keep trying to avoid. If you want to become a better runner, you can read about running form. You can learn about shoes. You can study heart-rate zones. You can watch someone else run. Some of that will help. None of it replaces the act of putting your body on the road and learning what happens when you actually move.
AI works the same way. You can take an introductory class. You can read a guide. You can watch a demo. Those things may help you understand the vocabulary. They may make the first few minutes less intimidating. But they do not build the muscle. The muscle is built when you use the tool on real work, get an imperfect answer, respond to that imperfection, and try again.
That is the whole game.
Use AI. Put in some bad prompts. Get some weird answers. Work with it.
Why are bad prompts and weird answers part of the process?
Bad prompts and weird answers are how the practice loop teaches you. A weird answer shows what the tool thought you meant. A bad output shows what context you forgot to include. Working through those misses is how a person learns to give better direction and develop judgment.
The bad prompts matter. The weird answers matter. They are not signs that the process failed. They are the process. A weird answer shows you what the tool thought you meant. A bad output shows you what context you forgot to include. A generic response shows you where your request was too broad. A confident but wrong answer reminds you that your judgment still has to stay in the loop.
That feedback is the training.
The mistake I see people make is that they want to skip straight to competence. They want the perfect prompt list. They want the perfect class. They want somebody to tell them exactly what to type so they never have to experience the awkward part. But the awkward part is where the learning lives.
If you never write a bad prompt, you never learn why it was bad. If you never get a weird answer, you never learn how to diagnose one. If you never push back on the tool, you never learn what kind of direction it needs. If you never use AI on a real task, you never learn where it fits into your actual work.
That is why a class can only take you so far.
A class can introduce the equipment. Practice teaches you how to use it under pressure. The fastest way to close the gap is to stop waiting for AI to feel finished and start using it on the work in front of you today.
What does the research say about AI fluency?
The research backs up the practice frame. AI-fluent workers learn through repeated, hands-on use, not by studying the subject from the outside. The pattern shows up in workplace data, productivity studies, and AI literacy frameworks alike.
There is research behind this, but the research is not complicated. Harvard Business Impact, citing work from Harvard Business Publishing Corporate Learning and Degreed, found that AI-fluent workers were much more likely to have learned through experimentation. The same work described AI-fluent employees as people who use generative AI frequently and understand its capabilities. That is an important distinction. Fluency is not awareness. Fluency is repeated use.
The numbers point in the same direction. Harvard Business Impact reported that AI-fluent respondents were two times more likely than other respondents to say they learned about generative AI through experimentation. Among AI-fluent respondents, 81 percent said AI made them more productive, 54 percent said it made them more creative, and 53 percent said it made them better prepared to solve complex business challenges.
That is what practice does. It turns a tool from something abstract into something you can actually use.
MIT Sloan covered similar evidence from research on generative AI at work. Danielle Li, Erik Brynjolfsson, and Lindsey Raymond studied a real customer-support environment, not a classroom exercise. The study included 3 million chats involving 5,179 workers, including 1.2 million chats from 1,636 workers after access to a generative AI tool. Access to the tool increased productivity by about 14 percent on average.
The detail I care about most is not only the productivity number. It is that workers accepted about 38 percent of the AI tool's suggestions on average. In other words, the productive users were not simply obeying the machine. They were working with it. They were accepting some suggestions, rejecting others, and learning how the tool fit into the job.
That is the muscle.
AI fluency is not the ability to press a button and accept whatever comes back. It is the ability to know what to ask, recognize when the answer is useful, spot when it is wrong, and keep shaping the output until it becomes something you can use.
Stanford Teaching Commons frames AI literacy in a practical way as well. Its AI literacy guidance includes using common generative AI tools, practicing basic and more complex prompting, understanding capabilities and limitations, and applying those ideas in real scenarios. That is a practice standard. It does not treat AI literacy as something you passively absorb. It treats it as something you demonstrate by using the tool and evaluating what happens.
That is the difference between knowing about AI and getting better at AI.
Knowing about AI is easy to fake. You can learn the terms. You can repeat the headlines. You can say model, prompt, context window, hallucination, agent, workflow, and automation in the right order. That may make you sound current in a meeting. It does not make you useful with the tool.
Getting better at AI is harder to fake because the work exposes you. You ask for something. The answer comes back. It is either useful, partially useful, or useless. Then you have to decide what to do next. Add context. Narrow the task. change the format. correct the assumption. ask for options. ask for a shorter version. ask for a more specific version. compare it against what you know. keep the good parts. throw away the bad parts.
That loop is where skill develops.
The more you do it, the more you start to understand the shape of the interaction. You learn that a vague request produces a vague answer. You learn that context changes everything. You learn that examples help. You learn that the first answer is often raw material, not the final product. You learn that the tool can be extremely helpful and still need supervision. You learn that the best results usually come after the second or third exchange, not the first one.
That is why I said the more you use the intelligence, the better you and it gets together.
I am not saying the model magically becomes a private version of itself that only understands you. I am saying the working relationship improves because you improve. You become better at giving direction. You become better at identifying the kind of work AI can accelerate. You become better at noticing when the answer is off. You become better at turning a rough output into a useful one.
The tool may be the same tool everyone else has, but your ability to work with it changes.
How should real estate agents practice using AI?
Real estate agents should practice on real, low-risk work they already need to complete. The fastest path from awkward to fluent is to use AI on the work that already shows up in the day, not on novelty exercises. Drafting client updates, organizing showing notes, and preparing inspection questions are where the practice loop pays off quickly.
That distinction matters for real estate because agents do not need AI as a novelty. They need it in the middle of actual work. A real estate agent does not become AI-fluent by asking a chatbot a random question once a week. An agent becomes AI-fluent by using AI on the work that already shows up in the day. AI fluency becomes more valuable when paired with real operating knowledge, which is why agents who already know their market and clients pick up the practice loop faster than newcomers.
Take a client update. The weak version is asking AI to write a generic message and sending whatever comes back. The useful version is giving the tool the real context: who the client is, what happened, what needs to be communicated, what tone is appropriate, what cannot be overstated, and what the next action should be. Then the agent reads the output and fixes it.
Take showing notes. The weak version is asking for a summary without enough information. The useful version is feeding in the actual notes, asking for themes, asking what follow-up questions a buyer may have, asking what items need confirmation, and then checking the output against the agent's own understanding.
Take inspection questions. The weak version is asking AI to tell you what to do. The useful version is using AI to organize the questions, separate urgent items from clarifying items, identify what needs a professional answer, and help prepare a cleaner client conversation.
None of that requires a perfect class. It requires reps.
This is where the running comparison is useful because nobody expects running to become comfortable before they run. The first run may be awkward. The pace may be wrong. The breathing may be rough. The route may be too long. That does not mean running does not work. It means the person is learning.
AI is the same. The first few prompts may be clumsy. The first few answers may be strange. The first few attempts may feel slower than just doing the task yourself. That does not mean AI is useless. It means you are still learning how to work with it.
The danger is quitting during that phase.
A lot of people try AI once, get a bad answer, and decide the tool is overhyped. That is like running once, getting tired, and deciding exercise is fake. The conclusion is wrong because the sample size is absurd. One bad prompt tells you almost nothing except that one prompt was bad.
The better conclusion is more useful. What did the tool misunderstand? What did I fail to provide? What did I assume it knew? What format did I need? What would I ask differently next time?
That is how you build calibration.
What is AI calibration?
AI calibration is the practical judgment a person develops by using AI repeatedly. It is knowing what to ask, what context to provide, when to trust an answer, when to challenge it, and how to turn rough output into usable work. Calibration comes from reps, not from passive exposure to the topic.
Calibration is the real skill people are looking for when they ask how to get better at AI. They do not just want to know which button to press. They want to know when to trust the answer, when to challenge it, when to ask again, when to stop, and when to do the work themselves. That judgment does not come from a certificate. It comes from repeated exposure to the tool's strengths and weaknesses.
The class might explain what calibration means. The reps build it.
So the practical answer is simple. Pick one real task today. Not a fake exercise. Not a demo prompt. A real thing you already need to do. Ask AI for help with the first version. Read the answer. Find what is wrong. Tell it what is wrong. Add the missing context. Ask again. Keep the useful parts. Throw away the rest.
Then repeat that tomorrow with another task.
Do that for client communication. Do it for notes. Do it for a market explanation. Do it for a checklist. Do it for a follow-up message. Do it for organizing your own thinking before a call. Keep the stakes low at first, but make the task real. Real work teaches you faster than practice theater. The pattern matches what I have seen for years in real estate, where successful agents build advantages through experimentation on real, repeated work rather than through one-time training.
Over time, the blank box stops feeling blank. You start to see the next prompt. You start to know what context matters. You start to recognize the kind of answer that is almost right but not quite. You start to feel when the model is making assumptions. You start to know how to steer.
That is not magic. That is practice.
And it is available to anyone willing to put in the reps.
AI is a muscle you build, not a class you take. If you want to get better at AI, use AI. Put in bad prompts. Get weird answers. Work with it. The more you use the intelligence, the better you and it gets together.
Judd Hoffman is CEO and Co-Founder of Ethica AI, building AI-powered tools for real estate transaction workflows.