Source: raw/Full_Claude_Code_Tutorial_for_Non-Technical_Beginners_in_2026_step-by-step.md (long tutorial, six builds); raw/From_Zero_to_Claude_Code_in_19_Minutes_no_code.md (19-minute condensed version); field note from raw/Claude_Code_-_Build_Your_First_AI_Agent_in_20_Minutes_100%No_Coding.md (Grace Leung, 2026-10-30); AI Master variant from raw/Ultimate_Claude_Code_Tutorial-_How_to_Use_Claude_Code_for_Beginners_in_2026.md (late September 2026)

Creator: Futurepedia (YouTube @futurepedia_io) URL: https://www.youtube.com/watch?v=bqJzIWAEn40 (full tutorial) · https://www.youtube.com/watch?v=5tgHMa2Z3jc (19-minute version) Duration: not in the transcript header for the full tutorial (about 8,500 spoken words); about 19 minutes for the short version, per its title Platform: YouTube Published: 2026-04-30 (full tutorial) · 2026-06-15 (19-minute version)

Futurepedia’s two beginner Claude Code tutorials walk a non-programmer from an empty folder to a live URL inside the Claude desktop app’s Code tab, never opening the code. The full tutorial does six builds: a browser game, a link-in-bio page, a product landing page, a Chrome extension, a Granola-to-Asana task board through connectors, and a calorie-tracking web app that calls the Claude API. The June video repeats the same loop more briefly. Both end on GitHub plus Vercel deployment. This article keeps the loop, the habits and the deployment steps, and marks what has changed since they were recorded.

Recorded April and June 2026

Both videos show the desktop app with three tabs: Chat, Cowork and Code. Since September 2026, chat and Cowork are one product on Pro and Max (see Claude Cowork, One Claude). The Code tab is still where desktop Claude Code runs as of October 2026 (see Claude Mods). The model advice is deliberately version-free (“the number at the end will change”). Read it against the current line-up: Opus 5.5, Sonnet 5.5 and Fable 5.1 (see Which Claude Model for Which Job).

Field note (2026-10-30): the same Code tab, used to build an agent

Grace Leung’s 2026-10-30 video (raw/Claude_Code_-_Build_Your_First_AI_Agent_in_20_Minutes_100%_No_Coding.md, Anthropic-sponsored) uses the desktop Code tab for an agent instead of an app. She starts local in auto permission mode on Opus 5.5, has Claude write CLAUDE.md and a brand-context file, adds an agent, a playbook skill and a memory file, then pushes the folder to a private GitHub repo and runs the agent from the mobile app and on a cloud schedule. Same GitHub step as Futurepedia’s, but used to run cloud sessions instead of deploying to Vercel. See Build Your First AI Agent in Claude Code Without Coding.

Key Takeaways

  • The core loop is plan, build, iterate. Pick or create a project folder, switch to Plan mode, send a short idea, answer Claude’s clarifying questions, then review the plan. You can add notes onto the plan before you click accept. After the first build, leave Plan mode off for ordinary edits and turn it back on for a big new feature.
  • One change per prompt. Futurepedia’s main iteration advice is to send changes one at a time. Several requests in one message are less likely to all land, and if something breaks you know which change caused it.
  • Screenshots do the debugging. When an error or a layout problem is easier to show than describe, screenshot it and drop it in the chat. This fixed the misplaced Chrome-extension button, the missing profile photo, the oversized-image API error and the Vercel filename error in the demos.
  • Asking a question stops the building. If you ask how something works, Claude explains and doesn’t build. That is how a non-coder learns the game logic without reading code, then asks for the next change.
  • Claude tests its own work. The demos show Claude opening the preview, clicking through the game, finding a wrong parameter, correcting course and removing its debug code before handing the result back. It also built a separate mobile layout for the link-in-bio page without being asked.
  • Put assets in the project folder, not the chat. Drop images and logos into the folder and name the file in the prompt. The landing page was built from a single futurefuel-can.png, and Claude wrote all the copy from it.
  • Context handoff takes two steps. First run /init to write a CLAUDE.md with the long-term facts (rules, stack, architecture), and ask Claude to update it after significant changes. Second, before you start a fresh session, ask for a “where we left off” summary covering architecture, decisions, current state and what’s left, then paste it into the new session. CLAUDE.md doesn’t capture what you were in the middle of. This matches Anthropic’s context-management guidance: compaction loses detail, and a fresh session with a good handoff loses less.
  • Models: plan and first build on the top model, iterate on a cheaper one. On Pro, Futurepedia uses Opus (or “Opus or Fable” in the June video) for the plan and first build, then Sonnet for follow-up edits. On Max, he runs the top model throughout. The wiki’s current guidance agrees and adds one detail: a fresh Sonnet session is cheaper than switching models mid-thread, because a mid-thread switch rereads the history without the cache (see Two-Model Workflow).
  • Usage is a five-hour window plus a weekly cap. Both show under Settings → Usage. Futurepedia recommends Pro (100/month) for heavy building, and says the free plan doesn’t include Claude Code. Limits have changed since the videos: Opus 5.5 raised the five-hour limits and added a saved reset (see Claude Usage Limits Playbook).
  • Parallel sessions come later. The full tutorial runs three builds at once (link-in-bio, landing page, Chrome extension), and the session panel shows a yellow dot when one needs input. Futurepedia’s advice is to get comfortable with one session first.
  • Connectors are MCP with a friendlier name. In the app they’re under Customize → Connectors, and Browse connectors opens the full list. For each connector you can set every action to allowed, needs approval or blocked. Futurepedia recommends Context7 for every builder, because it gives Claude current framework documentation and cuts hallucinated APIs (see Context7). He also names Stripe for payment builds.
  • His own summary of the three layers: connectors (MCPs) link Claude to tools, skills give Claude a process to follow, and plugins bundle several skills, connectors and more. Compare Skills vs MCP vs Plugins.
  • Run a security review, then a “what else” check, before going public. Asking for a security check triggers the built-in security-review skill without a slash command (compare /security-review in Claude Security Plugin). Then ask: “Is there anything else I should be aware of before making this live for other people to access?” In the demo that question caught old-browser support, untested touch controls and no mid-game restart.
  • Keep API keys out of the chat. Your Claude API key is billed separately from your subscription, from a balance you load in the Anthropic Console. Have Claude create .env.local, then paste the key into that file yourself in the side-panel editor. The file is listed in .gitignore, so it never reaches GitHub. That also means you add the key again by hand in Vercel’s environment variables.
  • You set up deployment once. Ask Claude to check for Git, then to connect GitHub. Claude walks you through installing Homebrew, installing the GitHub CLI and running gh auth login (github.com, HTTPS, browser login). From then on, “create a new GitHub repository for this project and push all the current code” is a single prompt. Import the repo in Vercel and click Deploy. Vercel redeploys automatically on every push.

AI Master variant (late September 2026)

AI Master’s beginner tutorial (raw/Ultimate_Claude_Code_Tutorial_-_How_to_Use_Claude_Code_for_Beginners_in_2026.md, https://www.youtube.com/watch?v=WdZjoMi6W7Y, publish date not in the transcript) runs the same desktop Code-tab loop with four builds: a podcast landing page, a save-a-quote Chrome extension, a Python downloads-folder sorter and a single-file subscription tracker. It was recorded about a week after Opus 5.5 launched, when the model menu showed Opus 5.5 with Sonnet 5 beside it. That fits a recording just before Sonnet 5.5 shipped on 2026-09-28.^[inferred] The video carries an ad for the creator’s own AI Master platform. What it adds to the Futurepedia loop:

  • Opus plans, Sonnet builds. Use Opus (the default on Pro) to write the plan and file list before any file exists, “the cheapest moment to change your mind.” Then switch the menu to Sonnet and say “follow the plan exactly and build it.” Skip the Opus step for small, clear scopes such as the Chrome extension and the tracker. “Knowing when not to use the expensive model is half of the pattern.” See Model vs Effort in Claude Code.
  • One folder per project. Never point the app at Desktop or Downloads. Check the connection with “what files are in this project.”
  • Narrow edits in three parts: what the code does now, what you want instead, and what must stay exactly as it is. For example: “group quotes by website and nothing else changes, especially the context menu and the markdown button.” First ask Claude to “explain what happens after I click save quote” in plain English.
  • Auto mode off for beginners. With it off, Claude shows each command and waits. The creator’s approval rule: approve routine package installs, local servers and tests quickly. Read slowly anything that deletes, overwrites or pushes live. Use the stop button to interrupt. He turns auto mode on only for longer handoffs.
  • A four-part brief for multi-step handoffs: the goal, the constraints, what counts as finished, and what you don’t want added. Then stop steering and test each item in the brief when Claude reports done.
  • A four-line CLAUDE.md for the tracker: plain JavaScript with no frameworks, full words in function names, everything in one file, never add a library without asking. Have Claude draft it with /init, then add your rules.
  • Two long-session traps. Run /compact when a session grows for hours. After two failed fixes for the same bug, start a fresh session with the full error and the file name, because “a short, clean context almost always beats a long one full of failed attempts.”
  • Cloud parallel threads, in beta at recording. A “new projects” feature runs several handoffs in parallel cloud threads. The creator advises beginners to wait, since every parallel job is another result to review.
  • Pricing as stated: Pro about 100. The API route “sounds cheaper,” but a long Opus planning session “can cost a few dollars on its own,” so move to it only once you know your real usage.

Try It

  1. First build (30-60 minutes). In the desktop app’s Code tab, create Claude Code Projects/<project>, switch to Plan mode, and send a two-sentence idea for something you’d use. Answer the questions, add one note to the plan, accept it, and let it build. Then make three single-change edits, at least one of them from a screenshot.
  2. Lock in memory. Run /init, open the new CLAUDE.md, then use the handoff prompt: “Summarize everything important about this project, the architecture, decisions we made, current state, and what’s left to do so I can paste it into a new session.” Open a fresh session on the same folder and paste the answer in.
  3. Connector build. Connect one tool you already use, such as a meeting-notes app plus a task manager. Set its actions to require approval, then ask for a small internal tool that reads from one and writes to the other. The full tutorial’s version reads Granola meeting notes, pulls teammates from Asana, lets you drag assignees onto action items, and pushes the tasks back to Asana.
  4. Ship it. Prompt “Check if I have Git installed, if not, install it for me”, then “Connect to GitHub so you can push code on my behalf and walk me through anything that needs manual input.” Push the repo, sign up for Vercel with that GitHub account, import the repo and deploy. Run the security check and the “anything else” question before you share the URL.
  5. Phone “app” without the App Store. For a mobile-responsive web app, open the deployed URL on your phone, tap Share, then Add to Home Screen. The calorie app (Morsel) used the phone camera this way. Futurepedia points out this is not a native app, and building a native one takes many more steps.

Implementation

Tool/Service: Claude desktop app, Code tab (Claude Code); GitHub plus the GitHub CLI; Vercel; optionally Neon (database) and Vercel Blob (file storage) from Vercel’s Storage tab; Anthropic Console for API keys.

Setup: A paid Claude plan. Futurepedia says the free plan doesn’t include Claude Code. Git comes with newer Macs. Claude gives Windows-specific steps when you ask, and Windows may show a PowerShell permissions prompt. One-time GitHub CLI authentication runs in the terminal panel built into the app. Your Mac password won’t echo while you type it.

Cost: Futurepedia’s figures from April 2026: Pro 100/month. Vercel’s free tier covers personal projects. API calls draw from a Console balance you pay for separately (“a few dollars goes a long way” for personal use, per the creator).

Integration notes:

  • A live deploy can fail while local works. In the demo, Vercel turned spaces in a filename into dashes and the page broke. A screenshot of the error was enough for Claude to fix it.
  • Apps with an API key or a database need extra Vercel setup. Add the API key under Vercel’s environment variables, because .env.local never leaves your machine. Morsel also needed Neon for its food logs and Blob storage for the photos. Futurepedia kept that repo private because it holds a working key, and he says a public app with keys, user data or payments needs a deeper security pass than this demo got.
  • Preview-panel workaround (April 2026 build). When the preview panel errored, opening the HTML file manually from the panel’s open-file button loaded it.
  • A Claude chat in a second window is where Futurepedia asks side questions (“which tech-stack option should I pick?”) so they don’t fill the build session’s context.

Open Questions

  • Futurepedia says the free plan doesn’t include Claude Code, and quotes Pro at 100. This wasn’t re-checked against current plan pages for this article. Check Surfaces Decision Framework before quoting prices.
  • The April video calls the desktop update “slightly buggy” in the preview panel. Whether later desktop builds fixed this isn’t recorded.
  • Both videos treat the built-in security review as enough for a personal project. Neither shows what it misses on an app with authentication and stored user data.