You now have a working website and enough Claude Code experience to build the next thing. Pick one real problem, give Claude a narrow task, and review the result before adding more.
The guide covered 24 parts, from opening a terminal to publishing a site. Finishing the course matters less than being able to take one idea and turn it into something that works.
What you can do now
You can:
- describe a result and give Claude the files and constraints it needs
- review changes instead of accepting a polished answer on trust
- keep work recoverable through versions, copies, and small steps
- publish a project and improve it from real feedback
You do not need to write every line yourself, but you still need to explain what you want, review the result, and know when to ask for help.
That responsibility does not disappear when the tool becomes easier to use.
Pick the next project
Choose something small enough to finish this week. A finished narrow tool teaches you more than another half-built platform.
You could extend the website with a resources page or a small portfolio. You could automate a repeated spreadsheet cleanup, create a calculator for one work problem, or build a dashboard from a file you already maintain.
For a team project, the first useful improvement may be less visible: add project instructions, document the validation commands, or turn one repeated task into a reusable skill.
My filter would be simple:
- Is there one person who will use it?
- Can you explain the problem in two sentences?
- Can the first version be finished in a week?
- Can you check the result yourself?
If the answer to the last question is no, make the first version smaller.
Choose the interface from the work
Claude's product boundaries keep changing, so I would choose the interface from the files and actions the task needs.
| Use | Good starting point |
|---|---|
| Explore an idea, ask questions, or work through a decision | Claude Chat |
| Inspect a software project, run development tools, and review file changes | Claude Code |
| Hand off multi-step knowledge work across documents and connected tools | Claude Cowork |
These overlap. Claude Code is available beyond the terminal, and Cowork can also use files and tools. Choose based on the work you need done. The friendlier screen is secondary.
Cowork is available on desktop. Web and mobile are still in beta for Pro, Max, and Team plans, and for Enterprise when an admin enables it. If you want Cowork to reach local files, use your browser, or control your computer from web or mobile, the Claude Desktop app must stay open. Check Anthropic's current Cowork guide before choosing it for a workflow.
For software work, check the current Claude Code interfaces instead of treating this course's terminal setup as the only option.
Reuse what worked
When an instruction produces a good result, save the useful context. That may become a project rule, a template, or a small reusable skill.
Do not save every prompt. Keep the part that explains the decision, constraints, and check. Save enough context so you do not need to explain the same decision again next week.
Build the first part
You still need to decide what should exist, who needs it, and why. Claude can build from that decision, but it cannot make it for you.
Pick one small project you can finish this week. Write down who it is for, what it must do, and how you will check that it works. Then build the first part.
Sources
- Use Claude Cowork on web, desktop, and mobile, Anthropic. Supports current plan eligibility, beta status, cross-device sessions, and which local capabilities require the desktop app.
- Platforms and integrations, Anthropic. Lists the current CLI, desktop, IDE, web, mobile, and integration surfaces for Claude Code.
Get Personalized Help
Copy this prompt to ChatGPT, Claude, or your favorite AI assistant. Fill in your details and get guidance tailored to your specific situation.
I'm using https://ivanmisic.net/blog/ai-tools/where-to-go-from-here to choose one small project I can finish this week. My context: - Person and problem: [WHO WILL USE IT AND THE PROBLEM IN TWO SENTENCES] - Existing material: [PROJECT FILES, SPREADSHEET, DATA, NOTES, OR AN EMPTY FOLDER] - Experience and time: [WHAT I HAVE BUILT, AVAILABLE DAYS OR HOURS, AND WHERE I NEED HELP] - Definition of done: [ONE RESULT I CAN CHECK MYSELF] Ask for missing facts, then reject or shrink anything that cannot be finished and reviewed within the week. Give me: 1. one project choice and why it passes the article's four filters; 2. the best current interface for the work, based on the files and actions it needs rather than the friendliest screen; 3. a first task narrow enough for one session, naming the likely files, constraints, and visible result; 4. a check I can run or perform myself before asking for the next part; 5. a short sequence for the rest of the week, stopping at a useful first version. Check current official documentation before relying on Claude Chat, Claude Code, or Claude Cowork availability or capabilities. Link what you verify and state what remains uncertain. Do not save every prompt as a reusable rule, choose an interface only because it is newer, or let the tool decide who the project is for and why it should exist.