From prompt to production. Master the tools that turn your ideas into live products — Claude Code as a design partner, GitHub for version control, and Cloudflare for instant deployment.
Perceive → Plan → Act → Observe → Adjust. The fundamental cycle every AI agent follows.
A customer feedback classifier with smart routing — our first real agent in production.
Trust spectrum, transparency, personality, and turn-taking — the design layer of agents.
Two valid approaches to building AI agents — knowing when to use which.
You gave Claude Code a single prompt and got a working AI app — let's appreciate what just happened.
The AI agent on your machine — what it is, how it thinks, and how to direct it like a trusted design partner.
Claude Code is an AI agent that lives in your terminal. It can read your files, write code, and run commands — autonomously.
Reads your files
Decides approach
Writes & runs code
Checks results
Think of it as Claude's working memory — everything it can "see" at once.
Claude starts "forgetting" earlier parts of the conversation.
Like a whiteboard that gets erased from the top — oldest content disappears first to make room for new content.
The context window has a limit. Here's how to keep it healthy.
/compact — Summarize & free up spaceClaude compresses the conversation history while keeping key information. Use it after working for a while.
/context — Check usageSee how much of the context window is used. Like checking how much storage is left on your phone.
New task? Start a new conversation. Claude works best with focused sessions, not 10-topic mega-chats.
Instructions that Claude reads every time it starts a new conversation in your project.
Without CLAUDE.md, every new conversation starts from zero. Claude doesn't know your project's conventions, style, or rules.
"Build me a page" → generic Bootstrap page
"Build me a page" → matches your design system, uses your API, follows your naming conventions
Every project gets its own CLAUDE.md. Plus you have one for you.
~/.claude/CLAUDE.mdYour personal preferences across all projects. Private to you.
"Always respond in Arabic. I'm a UX designer. Use modern, clean UI."
CLAUDE.md, README.md, and many other files use Markdown — a simple way to format text.
Bold text and italic
Instructions for Claude
Project front page on GitHub
Notes, docs, plans
Ask Claude Code to plan first instead of jumping straight into coding.
You: "Add a results export feature"
Claude: immediately starts writing code…
Maybe it picks CSV when you wanted PDF. Or adds it in the wrong place. Now you have to undo and redo.
You: "Add a results export feature. Plan first."
Claude: "Here's my plan: 1) Add export button below summary 2) Export as PDF with themes + quotes 3) Include interview metadata…"
You review, adjust, then approve. Right solution the first time.
The quality of your prompt directly determines the quality of the result.
Claude guesses what "better" means. Results vary wildly.
Claude knows exactly what to change. Predictable results.
What you want + Where it goes + How it should look + Constraints
GitHub, Cloudflare & Deployment — getting your AI Interviewer onto the internet for the world to see.
A place to store, share, and deploy your code — like Google Drive for developers.
Your code is safe in the cloud. Even if your laptop breaks, your work is preserved.
Share with teammates. Work on the same project together without overwriting each other.
Push code → it auto-deploys to a live website. Production in seconds.
You don't need to memorize commands — Claude Code handles that. But understanding the concepts helps.
Save a version of your work with a description. Like "Save As" with a note about what changed.
Send your snapshots to GitHub. Your code is now backed up in the cloud.
Get the latest changes from GitHub. Useful when working from a different device or with a team.
A separate copy to try things without affecting the main version. Merge it back when ready.
Anyone can see your code. Great for portfolios & open-source.
Only you and invited collaborators. For work projects.
Let's get your code on GitHub. Follow these steps — we'll do it together.
Go to github.com → Sign up (free).
Tell Claude Code: "Set up GitHub authentication for me."
Tell Claude Code: "Create a public GitHub repo called ai-interviewer and push my code."
Visit github.com — your files should be there!
Free hosting that auto-deploys every time you push to GitHub.
your-project.pages.devFrom idea to live product — this is how real apps are shipped.
Describe what you want
Builds the code
Stores your code
Auto-deploys
Anyone can access
~15 minutes to set up the full pipeline.
Every update is just "push to GitHub" → live in seconds.
Now you have two copies of your project. Understanding the difference is important.
Your computer · localhost:8000
Cloudflare Pages · your-project.pages.dev
Same pattern as GitHub — create account, authenticate, and let Claude Code do the rest.
Go to dash.cloudflare.com → Sign up (free).
Tell Claude Code: "Set up Cloudflare authentication for me."
Tell Claude Code: "Deploy my project to Cloudflare Pages."
It creates the project, connects your repo, deploys, and gives you a live URL.
https://ai-interviewer-xxx.pages.dev
Open it on your phone
Share with a partner
Run an interview
What would you improve?
Let's step back and appreciate the journey from prompt to production.
Session 3 → You wrote a prompt describing an AI interviewer.
Claude Code → Built a complete web application from that prompt.
GitHub → Stores and version-controls your code.
Cloudflare → Deploys it to the internet automatically.
Anyone → Can now use your AI tool with just a link.
Use Claude Code to improve your deployed AI Interviewer. Push updates to GitHub — they'll auto-deploy.
Questions before we wrap up? Next: applying everything to a real Almosafer UX case study — User Interviews, Personas, UX Audit, and Usability Testing, all powered by Claude Code as your design partner.