Build Something Real
Watch It Act. Then Build.
Read-aloud narrates the unit in a natural voice, paragraph by paragraph. It needs an account.
In short
Tools like Replit, v0, Lovable, and Claude Artifacts let anyone build working software using natural language descriptions. You're not learning to code — you're learning to direct a builder. Context engineering is the core skill.
The Hook
For most of the history of computing, building software required years of training. You'd learn a programming language, then a framework, then how databases work, then how to deploy to the internet. The path from "I have an idea" to "it exists and works" could take months even for experienced developers.
That timeline just collapsed. In 2026, people with zero programming experience are building real, working applications — apps, websites, tools — in hours. Not toy projects. Real things that real people use.
MIT Technology Review named "generative coding" one of their 2026 Breakthrough Technologies. One quarter of the startups accepted into Y Combinator's Winter 2025 batch had codebases that were 95% generated by AI.
This unit is where you stop learning about AI and start building with it.
The tools have reached a point where describing what you want is enough to get something built. Not perfectly — but well enough to ship, test, and iterate.
Here's the lineup of tools that make this possible:
Replit (replit.com) — An AI-powered development environment with an "Agent" that builds An app that includes both the part users see (the interface) and the behind-the-scenes stuff like the server and database. Basically the whole thing, front to back. applications from natural language descriptions. You describe what you want, Agent plans the architecture, writes the code, handles the database, and deploys it to a live URL. You don't need to understand the code. You need to understand what you want.
v0 by Vercel (v0.dev) — Generates React and Next.js UI components from text descriptions. Describe a dashboard, a landing page, or a form, and it generates the code with a live preview. Best for front-end design and interfaces.
Lovable (lovable.dev) — Full-stack app builder similar to Replit Agent. Describe your app, and it creates a working version you can deploy. A journalist in Kenya used it to build a team management tool for a fintech company — no coding background.
Claude Artifacts (built into claude.ai) — Creates interactive HTML, code, and visualizations directly in the chat. Great for prototyping ideas quickly before building them in a full tool.
Cursor (cursor.com) — An AI-first code editor used by 7+ million developers. It integrates AI directly into the coding workflow — you describe what you want to change, and it modifies the code in context.
In Y Combinator's Winter 2025 batch — the most competitive startup accelerator in the world — 25% of accepted companies had codebases that were 95% AI-generated. These weren't demo apps or prototypes. They were venture-backed companies raising millions of dollars.
The founders weren't all engineers. Some were domain experts (doctors, teachers, salespeople) who understood a problem deeply and used Describing what you want to build in plain language and letting AI write the code for you. You focus on the idea, the AI handles the technical stuff. Collins Dictionary's Word of the Year 2025. to build solutions their industries needed. The competitive advantage wasn't technical skill — it was problem understanding combined with the ability to clearly describe what they needed AI to build.
This represents a fundamental shift: the bottleneck for building software moved from "can you code?" to "can you describe what needs to exist?"
The common thread: you describe what you want in natural language, and the AI translates that into working software. This is context engineering (Unit 07) applied to building. The better your description — the more specific, structured, and complete — the better the result.
Kevin Roose, a technology columnist at the New York Times who is not a programmer, built multiple personal apps this way. One photographs the contents of his fridge and suggests lunches. Another tracks his reading. He called this "software for one" — tools built for your exact needs, by you, in an afternoon.
Before committing to a full build tool like Replit or Lovable, A quick, rough first version of your idea built to test if it works. It doesn't need to be perfect, just good enough to try out. your idea in Claude Artifacts. Describe what you want, see an interactive preview instantly, and Repeating the cycle of describe, build, review, fix, repeat to improve your project. Nobody gets it right on the first try. on it. Once you're clear on what you actually want, move to a full tool for a deployable version. This saves time because you figure out your requirements before the AI writes a full codebase.
The key mindset shift is this: you're not learning to code. You're learning to direct a builder. Everything you've learned so far — pattern recognition, context engineering, evaluation, understanding what AI can and can't do — converges here. You provide the vision, the context, and the quality bar. The AI provides the implementation.
That said, things will break. The AI will misunderstand your intent, build the wrong thing, or introduce bugs. This is normal. The workflow is iterative: describe → build → review → refine → repeat. The people who build successfully aren't the ones who get it right on the first prompt. They're the ones who know how to describe what's wrong and guide the AI toward what they actually want.
Step 1: Go to claude.ai. Type:
PromptCreate an interactive quiz that asks 5 questions about AI concepts we've learned — pattern recognition, training data, prediction vs understanding, AI economics, and context engineering. Show the score at the end with a short explanation for each answer.
Watch it generate a working interactive artifact.
Step 2: Test it. Take the quiz yourself. Is it accurate? Is it engaging? What would you change?
Step 3: Iterate. Tell the AI what to fix:
PromptMake the questions harder. Add a timer. Change the color scheme to dark purple (#7c6ff7 accent) on dark background. Add a 'share your score' button.
Step 4: Now go to replit.com (create a free account). Try Replit Agent: describe a simple tool you'd actually use — a to-do list, a habit tracker, a flashcard app, a personal journal. Watch it plan, build, and Putting your app on the internet so anyone with the link can use it. Tools like Replit handle this for you automatically..
Step 5: Share the link. Your app is live on the internet. You just built and shipped software.
The ability to build software just became democratized. Not "democratized" as a buzzword — democratized as in, you, right now, with no coding background, can build a working application and put it on the internet.
This changes the equation for careers, entrepreneurship, and personal projects. Ideas that used to require hiring a developer — or spending months learning to code — can now be prototyped in an afternoon. The bottleneck isn't technical ability anymore. It's having a clear idea and knowing how to describe it.
It also means the skills you've built in this program — understanding how AI works, context engineering, evaluating output — aren't abstract. They're the exact skills you need to build effectively with these tools.
Ship Something
Build a real, working project using any AI building tool (Replit, Claude Artifacts, v0, Lovable). Ideas if you're stuck: a study timer, a personal budget tracker, a quiz for a class you're taking, a recommendation engine for a hobby, a tool that helps organize something in your life.
- Choose a real problem — It must solve a real problem or serve a real purpose for you or someone you know.
- Build it — Describe what you want and let the AI build the first version.
- Iterate at least 3 times — Your first version will not be your final one. Review, describe what's wrong, refine.
- Document your process — What you described, what the AI built, what you changed, and what the final result is.
- Share it — Get a live link and share it with someone.
- 1Tools like Replit, v0, Lovable, and Claude Artifacts let anyone build working software using natural language descriptions.
- 2You're not learning to code — you're learning to direct a builder. Context engineering is the core skill.
- 3The workflow is iterative: describe → build → review → refine → repeat. Getting it right on the first try is not the goal.
- 4The bottleneck is no longer technical ability — it's having a clear idea and knowing how to describe it well.
Type: Article Title: I Tried Vibe Coding to Build an App — Rest of World URL: https://restofworld.org/2025/vibe-coding-ai-app/ Description: A Kenyan fintech manager with zero coding background builds a team management tool using Lovable. Shows this isn't a Silicon Valley phenomenon — it's global.
| Type | Title | URL | Description |
|---|---|---|---|
| Article | MIT Tech Review, "Generative Coding" — 2026 Breakthrough Technology | technologyreview.com/2026/01/12/1130027/… | MIT names AI-assisted coding a breakthrough technology |
| Reference | Karpathy, original "vibe coding" post (Feb 2025) | x.com/karpathy/status/188… | The post that launched the vibe coding concept |
| Article | Wikipedia, "Vibe Coding" | en.wikipedia.org/wiki/Vibe_coding | History and context of vibe coding as a practice |
| Article | Rest of World, "I Tried Vibe Coding to Build an App" (2025) | restofworld.org/2025/vibe-coding-ai… | Non-programmer in Kenya builds a real app using AI |
| Tool | Replit | replit.com | AI-powered development environment with Agent for full-stack building |
| Tool | v0 by Vercel | v0.dev | Generate React UI components from text descriptions |
| Tool | Lovable | lovable.dev | Full-stack app builder from natural language |
| Tool | Claude Artifacts | claude.ai | Interactive prototyping directly in chat |
| Tool | Cursor | cursor.com | AI-first code editor for developers |
| Book | Mustafa Suleyman, The Coming Wave (2023) | — | How AI will transform creation, economics, and power |
Last updated: May 21, 2026. Video embeds and links were refreshed on this date; the build steps still work as written.
Marking a unit complete, the Prove It check and your place in the course all need an account. The reading stays free.