There are three levels of AI, and most people get stuck before they even reach level one. The good news: moving up is not about being technical. It's about taking one next step from wherever you are today.
Here's each level, how to tell if you're on it, and the single move that takes you to the next one. Find yourself, then do just the next thing.
Level 1: Chatbots (it answers you)
This is ChatGPT, Claude, Gemini. You type, it responds. It's brilliant with words: answering questions, writing, summarizing, translating. But it only reacts. Nothing happens until you ask.
You're here if: you've opened one of these and had a back-and-forth conversation.
Never started at all? Here's your entire level-one on-ramp. Open ChatGPT, and paste this in:
Your first prompt
"I've never really used AI before. I want to [describe one thing you'd love help with this week]. Ask me one question at a time to understand what I need, then help me do it."
The whole leap is one shift: from asking about a task to handing the task over. Here's how to actually make it:
- Change your access first. A free chatbot can only talk to you. To let it do the work, you need a version that connects to your apps, like ChatGPT Work, Claude Cowork, or a paid plan with connectors switched on. This is usually the real blocker, not your skill.
- Pick the smallest real task you already do. Not a big scary one. Something you repeat every week: cleaning up a spreadsheet, sorting your inbox, turning messy notes into a tidy doc.
- Connect the one tool that holds it. Point the AI at that single Google Sheet, Notion page, or inbox, and give it permission to read what's there.
- Hand over the whole task, not a question about it. Instead of "how do I organize this sheet," say "organize this sheet for me and show me what you changed."
- Become the reviewer, not the doer. Your new job is to check the result before it's final, not to do every step yourself. That mental switch is the real level-up.
Level 2: AI Assistants (it does the task for you)
This is when the same AI connects to your actual tools, your Google Sheets, your Slack, your Notion, your email, and does the work inside them instead of just describing it. Think ChatGPT Work or Claude Cowork. You're still in charge: you point it at a task, it carries it out, and you check the result.
You're here if: you've had AI actually build, edit, or organize something in one of your real tools, not just tell you how to.
A level-2 starting task
"Connect to my [Google Sheet / Notion / inbox]. Read what's there, then [organize it / summarize it / draft the next steps]. Show me what you changed before anything is final."
At Level 2 you give one task at a time. At Level 3 you give a whole goal and let AI break it into steps on its own. Here's the shift:
- Stop planning the steps yourself. Instead of feeding it one instruction, then the next, then the next, describe the finished outcome you want and let it work out the order.
- Use a tool built to run on its own. This is where Claude Code and ChatGPT's Codex come in. They can plan, build, test, and fix in one flow, instead of waiting on you for each move.
- Start with something low-stakes. A little tracker, a simple script, a one-page tool. Pick a first project where a mistake costs you nothing but a redo.
- Let it fail and fix itself. The whole magic of Level 3 is that it catches its own errors. Don't jump in to correct every step. Give it room to check its own work and try again.
- Judge the outcome, not each step. You're moving from reviewing every action to reviewing the finished result and deciding if it's good enough.
Level 3: Agentic AI (it runs on its own)
This is the big story of 2026, and it's where tools like Claude Code and ChatGPT's Codex live. You give it a goal, and it works autonomously, running a loop where it checks its own output, sees what worked, and replans what didn't, without you approving every step. Because it can catch and fix its own mistakes, it handles multi-step projects that would break a simpler assistant.
You're here if: you've given a system a full outcome, like "build me a working tool that does X," and it planned, built, tested, and fixed it end to end.
A level-3 goal (for a tool like Claude Code or ChatGPT's Codex)
"Build me a simple [tool / tracker / script] that does [what you want it to do]. Plan the steps, write it, test it, and fix any errors yourself, then show me the finished result."
Once you're here, going further isn't about a fancier tool. It's about trust and rhythm:
- Graduate from experiments to recurring work. Let it handle the things you do every single week, not just a fun one-off you built to see if it worked.
- Build a review habit, not a redo habit. Your value now is in reviewing and directing the work, not in quietly re-doing it yourself when you get nervous.
- Set firm guardrails on anything that matters. Keep a human eye on anything that touches money, private data, or messages sent on your behalf. Fast and confidently wrong is still wrong.
- Stack small wins. Better than yesterday is still the whole game, even up here.
So, what level are you aspiring to?
The real question was never "what level are you on." It's what level you're ready to reach next, and what you're leaving on the table by staying where you are. You don't have to leap to level three. You just have to take the next step, and then the one after that. Better than yesterday is the whole game.