Five levels of working with Claude — a self-assessment · Resource

This is a self-check, not an article. Working with Claude falls into five levels, from asking questions to systems that run unsupervised. Go through them in order, check the lines that describe you, and stop where you stop checking: that's your level. Under each level is the one move that opens the next.

The rule for checking a box is simple: check it if you do this today, not if you know it would be possible. How to turn the boxes into a score is at the end, in "How to read your score."

(Claude is Anthropic's AI assistant; you talk to it in writing, it answers, reads files, and at the higher levels it also carries out tasks on its own.)

Level 1 — curious

Check what describes your work with Claude today:

[ ] I open Claude, ask a question, get an answer, and close the tab.
[ ] I ask for help with an email, or to explain something I just read.
[ ] I treat Claude like a search engine that returns paragraphs instead of links.
[ ] Instead of typing out what's on a screenshot, I just paste the screenshot (Claude reads images).
[ ] I save maybe half an hour a day on small things.
[ ] I don't have a single project set up yet.

Most people stay on this level forever. It's genuinely useful, but it's still the floor.

The move to the next level: set up your first project. A project is a dedicated space in Claude for one topic — your business, a side venture, a recurring type of work. You drop in a few source documents and a short brief: who you are and how Claude should answer you. From then on, every conversation inside the project starts already "briefed on the topic."

Level 2 — beginner

Check it if this sounds like your work:

[ ] I have projects set up, and they're the backbone of how I work with Claude.
[ ] I open a new conversation in a project and ask about earlier decisions — Claude pulls them up and picks the thread up where I left it.
[ ] I rely on memory: Claude remembers my role, my preferences, decisions from weeks ago.
[ ] I have connectors wired to tools I already use — Slack, Google Drive, Gmail, Calendar — so Claude reaches for the thread or document itself.
[ ] I ask for finished files: spreadsheets with working formulas, slide decks, documents, PDFs ready to download and send.
[ ] I ask for a chart in the conversation when it explains something better than text.
[ ] I use the Microsoft Office add-ins — Claude works directly inside Excel, PowerPoint, and Word.
[ ] I save more than five hours a week.

The ceiling here is clear: Claude still doesn't do anything on your computer. You're still the one copying results into other programs and making the changes by hand.

Two limits worth remembering as you check boxes:

Memory itself works on every plan, including the free one; searching past conversations is a paid feature.
Visualizations live inside the conversation and don't save as files, unlike the documents Claude creates.

The move to the next level: stop forcing the chat window to do everything. Open the Claude desktop app and go into the co-work tab.

Level 3 — intermediate

Check it if this is already how you work:

[ ] I work in co-work: Claude has access to files on my computer and figures out on its own how to reach the result I described.
[ ] I give it a folder and one instruction ("sort by type, standardize the names, write a summary"), and the work is done by the time I come back.
[ ] I use file access — reading, editing, and creating real files, within the permissions I've granted.
[ ] I have saved skills — recipes for repeatable tasks that I wrote up once and now just call on.
[ ] I have scheduled tasks — Claude works on a rhythm I set (mornings, Mondays) while the computer is on with the app open.
[ ] I run the work from my phone paired with the computer.
[ ] I use Claude design — I describe a prototype, a deck, or a page in plain language, and Claude designs it, sticking to my visual identity.
[ ] I save more than ten hours a week.

Limits worth keeping in mind:

Co-work is a paid feature (any plan above free), runs in an isolated environment, and only reaches the folders you explicitly allow it into.
More than a hundred ready-made skills are already published — pick the source you download them from carefully.

The move to the next level: set up a fixed folder structure. An "about me" file, a templates folder, a projects folder, an output folder. Tell co-work: "always read the about-me file first, never touch the templates, drop finished work in the output folder." That structure turns Claude into a predictable collaborator.

Level 4 — advanced

Check it if this is your everyday way of working:

[ ] I use Claude Code — Claude for programming and operational tasks, which reads files and carries out entire sequences of steps.
[ ] I run several sessions in parallel (five at once, easily), each on its own, and come back to a handful of finished tasks at the same time.
[ ] I keep a Claude.md file with the start-up brief: who I am, what the project's goal is, what to never do.
[ ] I have a habit: when Claude makes a mistake, I say "add this to Claude.md so you don't do it again."
[ ] I use plan mode — Claude reads first, lays out a plan, and waits for approval instead of acting right away.
[ ] I use sub-agents — a Claude specialized for one task, each in its own context window.
[ ] I wire in external tools through MCP (and when a simpler command-line tool exists, I reach for that instead — it burns fewer tokens).
[ ] I run a verification loop: I give Claude a way to check its own work — open a preview, run a test, fix it, until everything works.
[ ] I watch the context window (the compressed summary of the older conversation history) and turn repeated instructions into my own shortcuts.

Most people get stuck here for one reason: they manually supervise all the parallel work and become the bottleneck themselves. That doesn't scale — it's babysitting.

Two terms that help when checking boxes:

The context window is how much text the model "sees" at once: instructions, files, the conversation so far.
A token is the chunk of text the model operates on — roughly a piece of a word, and what you pay for.

(You don't have to work in a terminal — the same engine runs in a tab inside the app.)

The move to the next level: find the most repeatable thing you do every week — and turn that into your first automation.

Level 5 — architect

Check it if this is already how your work runs:

[ ] I close the laptop, and the work keeps happening — systems run while I sleep.
[ ] I have routines: saved configurations running in Anthropic's cloud, triggered on a schedule, on call, or by an event (the computer can be off).
[ ] I have hooks — guardrails that block dangerous commands or let me know when a long task finishes.
[ ] I control sessions through channels — outside the terminal, from a chat app for instance.
[ ] Before I write anything, I check whether someone has already built a ready-made skill or MCP server, and adapt it instead.

The most important thing on the whole ladder: the obstacle at this level isn't technical, it's trust. Almost anyone can set up a routine in the cloud, but few do, because handing the wheel to a system that runs while you sleep feels risky.

How to start building that trust: like learning to drive — you don't start on the highway, you start in an empty parking lot. Pick a low-stakes task (a daily summary for yourself, sent nowhere) and watch it run for weeks without touching it. Once you trust those runs, you'll trust the next ten.

How to read your score

Your level is the highest of the five where you checked lines describing your work today, not the one you know you could reach. If you stop checking boxes partway through a level, your level is the previous one, fully checked.

Five levels in order: curious, beginner, intermediate, advanced, architect.
Four of the five transitions open with a specific move — the first project, the co-work tab, a fixed folder structure, the first automation.
The last transition is different. Going from level four to level five doesn't happen by flipping on a feature, but through trust built by repetition.

You don't have to climb all the way to the top. What's worth knowing is which step you're standing on and what the next one is — and to make exactly the one move written under your level.