Day 1

Role, Task, Context, Format

20 minutes

Good prompts start from structure, not magic phrases. This framework is the base layer for everything else you'll build this month.

1. Pick a real work task you'd normally ask Claude about. Something you did last week.
2. Write your prompt using all four elements:
   • Role: Who is Claude playing? ("You're a senior copywriter who writes for B2B SaaS founders")
   • Task: What do you want done, specifically? ("Rewrite this email so it gets a reply")
   • Context: What does Claude need to know? (The recipient, the goal, the history)
   • Format: What should the output look like? ("3 versions, max 100 words each, no greeting")
3. Compare this to how you usually prompt. Notice what changes.

Anthropic's official guide: Prompting best practices. Bookmark this. You'll come back to it.

Day 2

Follow-Ups and Iteration

15 minutes

Most people treat AI like a search engine. Ask once, get an answer, accept it. The real leverage is in iteration.

1. Take yesterday's prompt and its output.
2. Run three follow-ups on it:
   • "Make it sharper. Cut any line that doesn't earn its place."
   • "What are the weak points in your last answer? Fix them."
   • "Now rewrite it as if you were actually confident about this."
3. Notice the quality curve. The first draft is almost never the best one.

The single biggest unlock in Week 1 is learning that prompting is a conversation, not a vending machine.

Day 3

Load Your Personal Context

25 minutes

Claude doesn't know your business. It doesn't know your voice, your clients, your goals, or your past work. Every conversation that starts from scratch is wasted context.

1. Open Claude on the web or desktop. Create a new Project (or start a conversation you plan to reuse).
2. Add the following as context:
   • A one-paragraph overview of your business
   • Your top 3 client or customer types and what they care about
   • A short "voice guide" (5-10 lines about how you talk and write)
   • 2-3 pieces of your actual past work as examples
3. From now on, start your work conversations from this context, not from zero.

This alone puts you ahead of 90% of AI users. Most people re-explain their business in every new chat. You won't.

Day 4

Three Reusable Prompt Templates

30 minutes

Most of your AI time is spent re-explaining context. Templates kill that tax. You write them once and reuse them forever.

1. Identify your three most common AI use cases. For most founders it's some mix of:
   • Writing something client-facing (proposals, emails, LinkedIn posts)
   • Summarizing or processing information (meeting notes, research, docs)
   • Brainstorming or problem-solving (strategy, copy, product)
2. For each, write a template prompt with all four Day 1 elements baked in. Use placeholders where specifics change.
3. Save these three templates somewhere you can grab them quickly. Notes app, Raycast snippet, sticky note. Doesn't matter.

By Day 14 you'll be converting these templates into real Claude Skills. For now, paper templates are enough.

Day 5

Extended Thinking Mode

15 minutes

For complex tasks, you want Claude to think before answering, not just pattern-match. Extended thinking is the toggle for that.

1. Pick a genuinely hard task. Something you've been stuck on. A business decision, a thorny email, a strategy question.
2. Ask Claude once normally.
3. Then ask the same thing again with extended thinking enabled.
4. Compare the two answers. The second will be slower and meaningfully better on hard tasks.

Rule of thumb: turn extended thinking on when you'd be willing to wait 30 seconds for a better answer.

Day 6

Spot and Fix Claude's Mistakes

20 minutes

Claude can confidently tell you the wrong thing. Every top 1% user has a BS detector. You need one too.

1. Deliberately push Claude into a mistake zone. Ask about something very recent (last week's news), something very specific (a local business's hours), or something that requires math it shouldn't trust itself on.
2. Watch the shape of the answer. Does it hedge? Does it confabulate confidently? Does it cite a source that may or may not exist?
3. Learn the red flags:
   • Very specific numbers without a cited source
   • "According to recent reports..." with no actual source named
   • Lists of names, dates, or stats (the highest-hallucination zones)
   • Any claim about its own capabilities or its own model version

From today on, verify any factual claim that matters. Claude is a thinking partner, not an oracle.

Day 7

Week 1 Audit

15 minutes

Before you layer on more complexity, make sure the foundations are solid.

1. Review the work you did on Days 1-6.
2. Answer honestly:
   • Which of the four elements (Role, Task, Context, Format) do you still skip out of habit?
   • Do you actually iterate, or do you still accept the first answer?
   • Is your personal context loaded in your main Claude setup?
   • Do you have three reusable templates?
3. If any of those are weak, spend today strengthening them.

Week 1 is the one most people never actually finish. Finish it.

Day 8

What a Skill Actually Is

20 minutes

Everyone talks about prompts. Almost no creator explains that Claude can learn a whole workflow from a file you give it once. That's what Skills are. They're the thing that separates power users from prompters.

1. Read Anthropic's official overview: What are Skills?
2. Then read the announcement post: Introducing Agent Skills
3. In your own words, answer: what's the difference between a really good prompt and a Skill?

The short version: a Skill is a reusable set of instructions Claude loads automatically when a task matches. You write it once, it runs forever, and it works across Claude Chat, Cowork, and Claude Code.

Day 9

Install Claude Desktop

10 minutes

Skills live inside Claude. You need the desktop app to use them properly.

1. Go to claude.com/download and install the Claude Desktop app for your OS.
2. Sign in.
3. Check your plan. If you're on the free plan, that's fine for Weeks 1 and 2. You'll need to upgrade to Claude Pro ($20/month) before Day 15 to access Cowork in Week 3.

Don't upgrade yet if you're on free. Finish Week 2 first and see if the playbook is working for you. Then commit.

Day 10

Build Your First Skill by Conversation

30 minutes

You don't write code to build Skills. You describe a workflow to Claude in plain English and Claude builds the Skill file for you. This is the single most underrated feature in the entire Claude ecosystem.

1. Read the short support article first: How to create a skill with Claude through conversation
2. Pick a workflow you do every week. Writing a weekly client report. Building a content brief from a raw idea. Onboarding a new client. Quoting a new project. Any task you've done enough times to describe.
3. Open Claude and paste this prompt, filling in your workflow:

4. Answer Claude's follow-up questions honestly. Let it build the Skill for you.

The first time you do this, it feels like magic. That's because it basically is.

Day 11

Turn It On and Use It

20 minutes

Building the Skill is half the battle. Actually using it on real work is the other half.

1. Follow the instructions in Use Skills in Claude to enable your new Skill.
2. Give Claude a real task that should trigger the Skill.
3. Watch what happens. Claude should automatically apply the Skill's instructions. If it doesn't, your Skill description probably needs to be more specific about when to use it. Go back and tweak it.

You'll know the Skill is working when Claude produces output that matches your usual format, voice, and structure without you having to repeat yourself.

Day 12

Build Your Second Skill

30 minutes

One Skill is an accident. Two means you have a system.

1. Pick your second most common task. The one right after whatever you built Day 10's Skill for.
2. Run the same conversational Skill-building process from Day 10.
3. Enable it and test it.

If you're tempted to build five Skills today, don't. Depth beats width in Week 2. Two well-built Skills beat five half-built ones.

Day 13

Study the Pros

20 minutes

Seeing how Anthropic itself designs Skills teaches you patterns you won't invent on your own.

1. Read Anthropic's Complete Guide to Building Skills for Claude (PDF). This is the canonical reference.
2. Browse the awesome-claude-skills community list for more examples.
3. Note 2-3 patterns or ideas you want to steal for your own Skills.

By the end of today, your mental library of what Skills can do should be significantly bigger than when you started Week 2.

Day 14

Week 2 Audit

15 minutes

You're no longer just a user. You're a builder. Most AI users will never get this far. Take a second to see what you've done.

1. List your two Skills and what they do.
2. Estimate the time each saves you per use.
3. Multiply by how many times you expect to use them per week.
4. That number is your Week 2 ROI. Look at it. That's what a power user looks like.

Day 15

Meet Cowork

25 minutes

This is the step most AI users never take. Cowork is where Claude stops being a chat partner and becomes an agent. It reads your files, writes to your folders, runs tools, and completes multi-step tasks while you watch.

1. Read Get started with Claude Cowork in full.
2. Open Cowork inside Claude Desktop. Click around. Get familiar with the interface.
3. In one sentence, explain to yourself: what can Cowork do that regular Chat can't?

The answer is: Cowork can actually finish tasks. Chat describes how to finish tasks. That's the whole difference.

Day 16

Your First Cowork Task

30 minutes

The first time you watch Cowork read your files and produce a real deliverable, the plateau breaks. You stop seeing AI as a text box and start seeing it as a coworker.

1. Create a small, safe folder on your computer. Put 3-5 real files in it. (Docs, notes, screenshots, CSVs, anything relevant to your work.)
2. Open Cowork. Point it at the folder.
3. Give it a simple task. Examples:
   • "Read everything in this folder and give me a one-page summary"
   • "Organize these files by topic and rename them accordingly"
   • "Turn these raw notes into a clean meeting summary"
4. Watch it work. Actually watch it. Don't tab away.

This is the "oh shit" moment.

Day 17

Watch Cowork Invoke Your Skills

20 minutes

Your Week 2 Skills weren't just for Chat. Cowork automatically uses them when the task calls for them. This is the moment the whole progression clicks: Skills are the expertise, Cowork is the agent that uses them.

1. Make sure one of your Week 2 Skills is enabled.
2. Give Cowork a task that should trigger the Skill. (Match the task to what the Skill is good at.)
3. Watch for the moment Cowork invokes the Skill. You should see it referenced in the session output.

If you'd built this Skill in Week 2 and never taken Week 3, you'd never see this. Skills without agents are instructions on a shelf. Agents without Skills are assistants without expertise. You now have both. Most AI users have neither.

Day 18

Task Chains in One Session

30 minutes

Real work isn't one step. It's "do X, then based on X do Y, then based on Y do Z." Cowork handles chains natively.

1. Pick a 3-step task you do regularly. Something like: "research this company, draft a cold email based on what you found, then save both to a folder named after the company."
2. Give it to Cowork as a single prompt.
3. Watch it execute each step in order. Note what it does well and where it asks you for input.

You've just built a mini workflow without writing any code.

Day 19

Delegate a Real Deliverable

45 minutes

Saving 30 minutes is nice. Delegating an entire 2-hour task is transformative. This is the first day you'll actually feel what "top 1% AI user" means.

1. Pick a real task from your week. Something that normally takes you 1-2 hours. A piece of client work, a research project, a content brief, a process doc.
2. Give it to Cowork with full context (your Week 1 personal context plus any task-specific details).
3. Let it work. Resist the urge to micromanage. Course-correct only when it's clearly going off the rails.
4. Review the output. Give yourself an honest grade: would you have been happy with this if a junior employee produced it?

This is the day the math changes for your business. Save yourself 2 hours once and the rest is procrastination.

Day 20

The Three Types of Tasks to Delegate First

20 minutes

Not every task is a good agent task. The best delegations share a pattern. The worst waste your time.

1. Make a list of every recurring task on your plate. Weekly reports, client updates, outreach, research, content, admin.
2. Mark each as one of three categories:
   • Good for Cowork right now: clear inputs, clear outputs, low stakes
   • Good for Cowork with a Skill: has a specific format or voice that needs a Skill first
   • Not good for Cowork yet: requires real judgment, live human relationships, or info Cowork can't access
3. Start Week 4 with the "good for Cowork right now" tasks as your candidates for scheduling.

Day 21

Week 3 Audit

15 minutes

You now have agents working for you. Pause and acknowledge that. Most people who own AI tools never get this far.

1. How many tasks did you successfully delegate to Cowork this week?
2. Which one surprised you the most with how well it went?
3. Which one surprised you with how badly it went, and why?

Bring the best candidates from Day 20 into Week 4.

Day 22

From "I Run Claude" to "Claude Runs"

15 minutes

Week 3 you triggered Cowork manually. Week 4 Cowork runs on its own, on a schedule, while you do other things. This is the difference between having an agent and having an agent system.

1. Read the "Scheduled Tasks" section of Get started with Cowork.
2. Understand the critical constraint: scheduled tasks only run while your computer is awake and Claude Desktop is open. (Don't schedule anything that needs to run overnight while your laptop is closed.)
3. Pick the Week 3 task that went best. That's your candidate for scheduling.

Day 23

Your First Scheduled Task

30 minutes

A task that runs itself every Monday morning is worth more than a task you have to remember to run sometimes.

1. Take the candidate from Day 22 and set it up as a scheduled task in Cowork.
2. Configure it to run on a cadence that matches your week. (For a weekly report, maybe every Monday at 9am. For a daily summary, every evening at 6pm.)
3. Let it run for the first time. Review the output.
4. Adjust the task description if the first run wasn't quite right, then let it run again.

You've just automated a recurring task. That's a legitimate system.

Day 24

Claude Dispatch

20 minutes

Assign from your phone, pick up from your desktop. You stop being tied to one device to work with your agents.

1. Read the Dispatch overview: Put Claude to work on your computer
2. Try it: start a task from the Claude mobile app and pick up the finished result on your desktop later in the day.
3. Note the kinds of tasks where this is most useful. (Long-running research, background drafts, any task you want running while you're in a meeting.)

Day 25

The Persistent Agent Thread

20 minutes

When tasks run for hours, you need a way to check on them without starting over. Persistent threads are the management layer.

1. Leave a Cowork task running from your desktop.
2. Check its status from the Claude mobile app or from a different device.
3. Confirm you can see progress, intervene if needed, and review output later without losing context.

You now have agents you can supervise from anywhere.

Day 26

Layer in Your Week 2 Skills

30 minutes

Scheduled tasks plus Skills equals agents with real expertise. This is the actual "top 1% user" stack.

1. Take one of your scheduled tasks from Day 23.
2. Make sure a relevant Week 2 Skill is enabled.
3. Let the task run again. Compare the output to the pre-Skill version.
4. If the quality jumped, you just proved the full progression works: Prompts, then Skills, then Agents, then Agent Systems. Every layer compounds.

Day 27

Stress-Test Your System

30 minutes

Things break. You need to know HOW they break so you can fix them before they bite you in front of a client.

1. Deliberately set up a task that's likely to fail in a recoverable way. Give it bad input, incomplete instructions, or a file it can't find.
2. Watch what Cowork does. Does it ask for clarification? Does it guess? Does it stop and flag the problem?
3. Based on what you learn, add a guardrail to your scheduled tasks. (A clearer input spec. A "if you're unsure, stop and ask" instruction. A fallback file.)

The difference between hobbyists and operators: operators know their system's failure modes.

Day 28

Week 4 Audit

15 minutes

You've built an autonomous AI system. Not a theoretical one. A real one that runs in your account right now.

1. List every scheduled task you've set up.
2. Estimate the weekly time saved by each.
3. Add them up.
4. Look at that number. That's what the top 1% looks like.

Day 29

Your Personal AI System Audit

45 minutes

You now have a real AI system. Not a theory. A system. Time to look at it from 10,000 feet so you know what you have and what's missing.

1. List everything you've built in 30 days:
   • Prompt templates (from Day 4)
   • Your Week 1 personal context (from Day 3)
   • Claude Skills you created (from Days 10-12)
   • Cowork workflows you use manually (from Week 3)
   • Scheduled tasks (from Day 23)
   • Dispatch and mobile workflows (from Days 24-25)
2. For each, note how much time it saves you per week, how mature it is (brittle vs solid), and what would make it better.
3. Pick your weakest link and your highest-leverage next improvement. Those are your two priorities for the week after graduation.

You now have something almost nobody else has: a documented inventory of your own AI system.

Day 30

Where To Go From Here

30 minutes

This isn't the end of anything. You're at the starting line of actually being in the top 1% of AI users. Most people quit at Day 7. A few quit at Day 21. If you're reading this, you didn't quit, and that puts you in rare company.

From here, the path splits. Pick the direction that fits you:

1. Go deeper in the same stack. Add 2-3 more Skills. Schedule 2-3 more tasks. Double your time savings in the next 30 days. This is the right move for 80% of readers.
2. Teach someone else. The fastest way to lock in what you've learned is to walk another founder through Week 1. Start with the one that asks you the most AI questions.
3. Cross the Managed Agents threshold. If you want to build agent workflows you can sell, read the bonus section above and start with the Managed Agents quickstart. This is the technical frontier.
4. Turn it into consulting. You now know more about practical AI than 99% of the people trying to hire AI help. Offer what you know. Start with one person you know who's stuck at the prompts plateau.

Your final action: pick one of the four paths. Write it down. Commit to one next step you'll take this week.

That's the whole thing. You went from ChatGPT user to agent operator in 30 days. Now go build.