1
Level 1
Clueless
You've heard of AI but haven't tried it, or you tried it once and got a bad result and gave up. You think it's either magic or useless — you're not sure which.
Where you are
Never used Claude or ChatGPT seriously. Maybe saw a demo. Think it's for tech people or students who want to cheat.
The one thing to do next
Go to claude.ai, sign up for free, and ask it one real question from your actual life. Just one. See what happens.
Your first prompt — try this
I'm new to AI. I work as a [your job] and my biggest daily frustration is [describe it]. Can you help me with this?
Try it on Claude free →
2
Level 2
Curious
You've tried it a few times. You got some useful answers but also some that felt generic or wrong. You're not sure when to trust it or how to get better results.
What you've learned
Claude can write things and answer questions. It sometimes hallucinates. It's not always right. You use it occasionally but not habitually.
The bottleneck
You're asking one-line questions and getting one-size-fits-all answers. You haven't yet understood that context is everything.
The upgrade — add context
Instead of: "How do I ask for a raise?"
Try: "I'm a 32-year-old accountant at a mid-size firm in Johannesburg. I've been in this role 2 years, I've taken on additional responsibilities, and my market research shows I'm 15% below market rate. Help me build the case and script the conversation."
Learn this in Prompt School →
3
Level 3
User
You use Claude regularly for specific tasks — mostly writing, summarising, or answering questions. It's become a useful tool but you're still using it like a search engine.
What you're doing well
You've found a handful of use cases that work. You trust it for drafting emails, summarising documents, explaining concepts. You use it weekly.
What you're missing
You stop at the first answer. You don't push back, go deeper, or ask follow-up questions. The conversation ends where it should begin.
The upgrade — use dialogue
After Claude gives you an answer, ask:
"What's the strongest argument against what you just said?"
"What am I probably not thinking about?"
"Go deeper on point 2 — give me specifics."
See Lesson 3: Push Back →
4
Level 4
Prompter
You've learned that how you ask matters as much as what you ask. You give context, specify formats, and assign roles. You get noticeably better results than most people around you.
Your toolkit
You use context, format instructions, role framing, and constraints. You know the difference between a weak prompt and a strong one. You've done Prompt School or equivalent.
The next frontier
You still work one task at a time. You haven't yet built reusable prompts, system instructions, or multi-step workflows that save time every day.
The upgrade — build reusable prompts
Pick the 3 tasks you do most often at work. For each one, write a master prompt template with placeholders: [CLIENT NAME], [SITUATION], [DESIRED OUTCOME]. Save them somewhere. Run them every time.
Browse 60+ starter prompts →
5
Level 5
Practitioner
AI is now genuinely integrated into your daily work. You have saved prompt templates, you use Claude for complex multi-step tasks, and you save multiple hours per week consistently.
What this looks like
You have a personal prompt library. You use Claude for research, drafting, analysis, and decision support. You're noticeably more productive than colleagues who don't use AI.
What to learn next
System prompts and custom instructions. How to give Claude a persistent persona and set of rules that applies to all your conversations in a project.
The upgrade — system prompts
In Claude, create a Project and add instructions like:
"You are my personal finance advisor. I am a 35-year-old South African earning R45,000/month. Always use ZAR. Always reference SARS rules. Always be direct and practical — no generic advice."
Set up a Claude Project →
6
Level 6
Integrator
You've connected Claude to other tools — documents, spreadsheets, APIs, or workflows. AI isn't just a chat window anymore; it's embedded in how you and your team work.
What this looks like
You use Claude via API for at least one workflow. You've built or used integrations with tools like Notion, Google Docs, or custom scripts. You understand tokens, costs, and model selection.
The key insight at this level
You've stopped thinking of AI as a chat tool and started thinking of it as an API that can reason. The bottleneck is now your ability to design systems, not to write prompts.
The upgrade — your first API call
Get an API key from console.anthropic.com. Use the Anthropic Python or JS SDK to send one message programmatically. That's it. Once you've done it once, the rest follows.
Anthropic API docs →
7
Level 7
Designer
You design AI-powered products and workflows for others to use. You think in terms of systems, user experience, and reliability — not just prompts. You ship things built with Claude.
What this looks like
You've built internal tools, client-facing products, or automated workflows using the API. You design the system prompt, the user experience, the error handling, and the evaluation criteria.
Your mindset shift
You no longer ask "what can Claude do?" You ask "what problem needs solving, and is AI the right tool?" You know when not to use AI as well as when to use it.
The upgrade — evaluate your outputs
Build an evaluation set for your most important AI workflow: 20 test inputs with expected outputs. Run your prompts against them. Measure pass rate. Iterate. This is how production AI systems are built.
See what Claude-built products look like →
8
Level 8
Builder
You build AI-native products, teams, or businesses. You use agents, tool use, multi-step reasoning, and evaluation frameworks. You're not just using Claude — you're building with it at scale.
What this looks like
You use Claude's API with tool use, computer use, or agent frameworks. You've shipped products used by others. You think about latency, cost, reliability, and model drift the same way a software engineer thinks about performance.
The honest truth
Most people will never reach Level 8 — and don't need to. Levels 4 and 5 already put you in the top 5% of AI users. The ladder exists so you know what's possible, not to make you feel behind.
Where builders go from here
Anthropic's Claude Agent SDK, tool use documentation, and the Model Context Protocol (MCP) are the current frontier. The community is small and the opportunities are large.
Anthropic agent docs →