Today I’m sharing a guest post on how to strengthen your AI skills. It’s by Frank Andrade, an AI instructor who has helped thousands of people on Substack master ChatGPT and Claude with beginner-friendly guides and in-depth tutorials.
Frank: The gap between AI novices and experts is growing.
One group is saving 5 minutes on emails.
Others are running systems that work while they sleep.
The good news? You can progress faster than you might think.
The Five Stages:
Stage 1: AI Beginner
Stage 2: Prompt Crafter ⬅️ (where many people stall)
Stage 3: AI Literate
Stage 4: Workflow Builder
Stage 5: System Builder
Regardless of which stage you’re at, the resources in each section below will help you sharpen your AI skills.
New to Wonder Tools? Take a look at the most popular past posts
Stage 1: AI Beginner
This is where everyone starts.
You have a ChatGPT, Gemini, or Claude account. You use it like a fancier Google. You ask one-off questions. Maybe you ask it to help write a quick email or generate a few images. That’s about it.
It’s useful, but it doesn’t feel like a superpower yet. Don’t worry. We’ll get there.
The first move at this stage is picking your tools.
AI tools fall into two categories:
General-purpose: Handle multiple tasks including text and visuals
Task-specific: Focus on a specific kind of content or media type
Frank’s Recommended Toolkit
Start Here
Worth Exploring
Gemini Notebook, Claude Cowork, and Claude Code
Specialized Tools
Gamma (presentations), Google’s Nano Banana Pro (images), ElevenLabs (audio), HeyGen (video), and Napkin (infographics)
The Big 3: Select a Starting Point
For beginners, the differences between ChatGPT, Gemini, and Claude aren’t significant. Still, here are some things to keep in mind:
ChatGPT: rich feature set, including voice and video modes and image generation
Gemini: works well with Google Docs, Sheets, Gmail, YouTube, etc.
Claude: strong at automation and building tools/apps; doesn’t generate photos or videos, though it does assist with design
If you’re a beginner, pick one to focus on. It’s fine to select what your company provides, what’s familiar, or one you’re curious about. Stick with it for a few weeks to see when it saves you time, how well it handles technical tasks, and where it’s most useful.
Frank’s current stack: Claude Cowork, and Claude Code for serious work, Gemini Notebook for learning, and ChatGPT for voice.
Your stack doesn’t have to look like mine. Build your own based on what you actually do.
Here are some of my resources for this exploratory stage:
Stage 2: Prompt Crafter
When it comes to AI chatbots, how you ask is as important as what you ask.
At this stage, get to know this basic prompt structure:
Instruction: what you want
Context: background the AI needs to know
Constraints: length, tone, format, things to avoid
Example: when possible, provide an example of what “good” looks like
For more detail, refer to this robust prompt formula shared by Anthropic, which makes Claude:
You’ll rarely need to use all the elements of the formula, but it’s good to know they exist. Think of them as a menu to pick from.
At this stage, you should also know a few prompt-improvement tricks. For example, when you finally land on a great answer after a long back-and-forth, ask the AI to write the prompt that would have gotten you there on the first try.
Here are some resources for this stage:
Who Needs to Go Beyond Stage 2?
Stage 2 was enough in 2023.
In 2026, AI has evolved and can run multiple tasks at once on its own. There’s nothing wrong with being in stage 1 or 2 if you’re a casual user who proofreads emails and generates images once in a while
But if you want to master AI, you need more.
That’s what we’ll cover in the next stages:
Stage 3: Develop AI literacy. Use it as a thinking partner.
Stage 4: Build AI workflows for the work you do repeatedly.
Stage 5: Build AI systems that run while you sleep.
Stages 3, 4, and 5 are for professionals, builders, and business leaders who want to turn AI into a competitive advantage.
Stage 3: AI Literate
Understanding how AI actually works changes how you use it.
People with AI fluency know when to trust AI outputs, when to push back, and how to frame problems to get more useful results. That’s what separates those who use AI as a novelty from those who deploy it for its practical advantages.
People at this level know these AI concepts:
Hallucinations: AI can confidently fabricate facts, stats, and citations. Always verify outputs when accuracy matters.
Context Window: AI assistants can remember a limited amount per conversation. Once that limit is exceeded, your AI forgets earlier context.
AI Sycophancy: AI models are trained to be helpful, sometimes to a fault. They may validate your perspective even when you’re wrong. They rarely challenge assumptions unless prompted.
Knowledge Cutoff Dates: Models generally have training cutoffs. That means they may not know about recent events. If an AI assistant has access to web search, it can gather new info.
Data Bias: AI inherits biases from training data. Be cautious when using AI for decisions about sensitive topics.
Privacy & Safety Basics: Know the data policies for your AI tool and your organization. Read about how to use private AI tools for free, and how to set up a private AI toolkit.
Having AI literacy means knowing how to manage AI limitations.
Example: Let’s say you’re a leader who wants to combat AI sycophancy to make better decisions. Create anti-sycophancy instructions that nudge an AI assistant to disagree with you constructively.
At this stage, move from “AI thinks for me” to “AI helps me think better.”
That means using AI as a sparring partner. Here are some ways to do that:
Ask AI to generate different solutions and grade them
“Generate three solutions. Score them from 1 to 10 and justify the score.”
Use AI to critique responses, opinions, plans, etc.
“Be my adversarial friend. Argue the strongest case against Option A. List five reasons why it might fail and the early warning signs. Suggest five ways to reduce those risks.”
Generate questions to refine outputs
“Before drafting anything, ask the five questions that would most alter the outcome of this memo: [1-sentence purpose].”
Resources for this stage:
Stage 4: Workflow Builder
The trick at this stage is designing your own workflows, not using more tools.
A workflow is a multi-step process where AI plays a role at each step. You might still drive it by clicking, uploading, and reviewing. But you don’t start from scratch every time.
Here’s a workflow I built:
Learn a foreign language with AI
Step 1: I answer questions about my learning preferences, challenges, and goals
Step 2: AI generates a “language DNA profile”
Step 3: I record myself speaking the language for five to 10 minutes → upload the transcript + my DNA profile → AI gives feedback tailored to me
This is a workflow rather than a full automation because I still trigger each step. I generate a profile. I upload the inputs. I hit go.
Here’s how to build your own AI workflow:
Pick one task you do repeatedly and turn it into a 3-step workflow
Pick one or two tools to help you optimize this workflow
Write a prompt for each step. If possible, add examples of what “good” looks like.
Test it with a real example and iterate until the output is what you need
Set up a Project (in ChatGPT or Claude) to keep your context, instructions, and reference files in one place
Resources for this stage:
For most professionals, this stage is more than enough. But if you want to multiply your output without working more hours, there’s one more level.
Stage 5: AI System Builder
Stage 4 was about workflows you run. Stage 5 is about systems that run themselves.
There’s a mindset shift here.
You stop asking, “How do I do this faster?” Instead, you frame the question as, “How do I build something that does this for me, even when I’m sleeping?”
Here’s a system I recently built:
The problem: I publish daily Substack notes. Many of my notes are inspired by viral content I find on Instagram, YouTube, and X. The old process took hours every day: saving links, watching videos, downloading images, writing hooks, etc.
The system I built (simplified):
Collection: My team and I save viral content to a shared group chat throughout the day
Trigger: At 3 a.m., an automation kicks in (I’m sleeping)
Processing: Claude reads the group chat, checks what’s new for today, then downloads everything (text, images, videos)
Generation: Each piece runs through my Substack notes generator (a separate workflow that uses my brand voice and note formats)
Output: When I wake up, I open my “Today” folder. Inside are 3 subfolders: YouTube, X, and Instagram. Each has media content + 5 draft notes ready to edit and post.
I pick one draft, edit it, and post it. What used to take two to three hours now takes minutes.
To build AI systems, you need tools that go beyond chatbots:
Claude Cowork: My favorite for delegating things to Claude without being technical. It runs in a sandboxed environment for safety, so you may run into limitations.
Claude Code: A more advanced version of Cowork. I use it whenever Cowork can’t build something for me. You don’t need to be a coder. You describe what you want, and it builds it. Caveat: it can feel scary at first.
n8n: Popular no-code tool for AI agents and multi-step systems. I don’t use it, but many builders do.
Resources for this stage:
Two Final Tips
Don’t skip stages. AI literacy and prompting may feel boring compared with building your own app, but they make every stage above them work better.
You don’t need to be at Stage 5 to get serious value. Many professionals get a 2x improvement just from reaching Stage 3.
Pick one stage above where you are right now and focus on it for the next few weeks. That’s how you’ll go from AI beginner to pro by the end of 2026.
Thanks to Frank Andrade for this guest post.
Wonder Tools Quick Survey
Help me understand who reads Wonder Tools and what matters to you. It takes about a minute (I timed it!). You’ll get my updated list of favorite free tools in an email within a few days of submitting it. Your responses and identifying info are confidential.
The jump from workflow builder to system builder changes where you get to check the work.
When you trigger a workflow yourself, you see what goes in and comes out.
Once it’s running while you sleep, you have to decide which mistakes can wait until morning and which ones should stop the system.
Getting something to run without you is only half of it. You also need to decide when it shouldn’t.
Turn your marketing meetings into finished work. Stages 4 and 5 are where most marketers stall. A notetaker leaves a list. A workflow turns the transcript into a brief, a follow-up draft, and next steps sitting ready for a human yes. The system that runs while you sleep only helps if approvals stay on the sensitive sends.