Key insights:
Business owners are making hiring, firing, and promotion decisions based on AI fluency. The gap between people who use AI well and those who don't is growing fast. This isn't about hype. It's about practical skills that directly affect your career and your output.
The good news? You don't need a computer science degree. You don't need to learn to code. You just need a structured approach and about three months of consistent practice. Ali Abdaal recently shared a five-phase framework for going from casual ChatGPT user to genuinely fluent. Let's break it down.
If you've tried ChatGPT a few times but haven't built it into your daily workflow, you're in the majority. Most people are still at this stage. The key difference between people who get real value from AI and those who don't isn't intelligence. It's habit.
The framework covered here takes you from zero to automated workflows in roughly three months. Each phase builds on the last. Skip a phase and you'll struggle with everything that follows.
AI fluency means you instinctively reach for AI tools whenever you're thinking, creating, or problem-solving. It means you have systems that get better over time. It means some of your repetitive work runs on autopilot.
That's the destination. Now let's walk through how to get there.
This works whether you're an employee, a freelancer, or a business owner. The examples come from a real team running a content business, but the principles apply to any role. If you do knowledge work of any kind, this framework is relevant to you.
You can also grab Ali's free AI learning guide to follow along step by step.
The first two weeks are about building habits and using AI as a thinking partner. These phases feel simple, but they create the foundation everything else depends on.
Week one is all about setup. Here are the five foundations:
These five things sound basic. They are. But most people haven't done them. If you skip this phase, every later phase becomes harder than it needs to be.
Week two is where AI starts adding real value. The key shift here is that you're not asking AI to do your work. You're asking it to help you think better about your work.
For example, you could tell Claude your role, your goals, and your biggest challenges, then ask it to coach you. You could say something like: "I want you to interview me about what I actually do in my role and help me identify what's high leverage and what's probably a waste of time."
This works for literally any job. The AI acts like a very smart colleague who has read a lot of books but has no context about your specific situation. You provide the context. It provides the questions and frameworks.
No, not blindly. Think of AI output as suggestions from a well-read intern. You still need to apply your own judgment. The value comes from the AI asking you questions you hadn't considered, not from following its advice as gospel.
By the end of week two, you should have the habit of turning to AI whenever you're stuck. Even when you're not stuck, you should be using it to pressure-test your thinking and optimize your approach.
This is where most people start, and that's the problem. Jumping straight to "write me an Instagram post" produces generic garbage. Phases one and two exist so that when you get here, you actually know how to give AI the context it needs.
This concept comes from Dan Martell's book Buy Back Your Time. The idea is simple:
You never want AI doing 100% of the work. That's how you end up with AI slop that nobody wants to read or use.
Instead of saying "give me 50 content ideas," a better approach is to provide transcripts, competitor examples, strategy documents, and specific constraints. The more context you give, the better the output.
Then you review the results, pick the ones you like, feed those back to the AI, and ask for more along the same lines. This iterative process is where the real magic happens. You might start with 20 ideas, like 5, then generate 50 more in that direction and like another 10. Now you have 15 solid ideas with minimal effort, but every single one has been vetted by a human.
Taste is the thing that separates people who use AI well from people who don't. You need an intuitive feel for what good looks like in your domain. When AI produces something and you feel that internal cringe, that's your taste telling you the output isn't good enough.
That cringe is valuable. It means your bar is higher than what AI produced. Your job is to give feedback, adjust, and iterate, just like you would with a junior team member. If you don't have taste yet because you're new to a field, that's okay. You'll develop it over time. But be aware that AI can't replace judgment you haven't built yet.
By this point, you're getting real value from AI. But every time you use it, you're starting from scratch. Phase four fixes that by turning your best prompts into reusable systems that improve over time.
Think of it like a recipe. The first time you bake a cake, you follow a basic recipe and it comes out okay. Over time, you tweak the sugar, add chocolate sauce earlier, adjust the timing. After dozens of iterations, you have a perfected recipe.
Prompt engineering works the same way. Your first prompt for generating content ideas might be basic. Then you notice the hooks are too generic, so you add "avoid generic advice." Then they're too long, so you add "keep each hook under 20 words." Then you notice rhetorical questions creeping in, so you add "never use rhetorical questions."
Each iteration becomes a new version. Over time, you build a library of battle-tested prompts for every recurring task in your work. You can use tools like TextExpander to create keyboard shortcuts that instantly expand your full prompts.
Once you have a systemized prompt library, you can start experimenting. Try the same prompt across ChatGPT, Claude, and Gemini. You'll quickly discover that certain models work better for certain tasks.
Maybe Claude is better for long-form writing. Maybe ChatGPT is better for brainstorming. Maybe Gemini is better for research. The point is that your prompt library gives you a consistent baseline to test against.
Not everything can be done through a text window. You might need AI tools for slide decks, image generation, video editing, or data analysis. The mistake people make is getting overwhelmed by the hundreds of new AI tools launching every week.
Don't worry about keeping up with everything. Find the specific tools that solve your specific problems. That's it.
This is where things get seriously powerful. Instead of talking to AI every time, you set up systems that run automatically in the background.
There are roughly four levels:
Most people don't need to go beyond level 2. Zapier alone can automate a huge amount of repetitive work.
Not everything should be automated. The discipline is deciding what's worth automating versus what's worth doing manually versus what you can just stop doing entirely.
A good rule of thumb: if you're doing the same manual task for several hours every week and it follows a predictable pattern, it's a strong candidate for automation. If it requires nuanced judgment every time, keep it manual but use AI to assist.
If you find yourself drawn to this world of automation, it's worth noting that automation development is becoming a real career path. The Complete RPA Bootcamp takes you from beginner to professional Automation Developer, covering Robotic Process Automation, Agentic Automation, and Enterprise Orchestration. Instead of worrying about AI replacing your job, you become the person building the automation. It's worth checking out if this space excites you.
Here's a concrete example. A team runs weekly coaching calls with students. Every call is automatically recorded and transcribed. A weekly automation pulls all transcripts, combines them with Slack support conversations and CRM data, then generates a summary report for each student covering their wins, struggles, and areas needing support.
That process used to take hours of manual admin every Friday. Now it runs automatically, freeing up coaches to spend more time actually helping students instead of compiling data.
This is the real promise of AI automation. Not replacing humans, but removing the tedious parts of work so humans can focus on the high-value stuff.
For a full walkthrough of all five phases with real examples and demonstrations, watch Ali Abdaal's video embedded below. He covers each phase in detail with specific prompts and use cases from his own team. It's one of the most practical AI learning resources available right now.
