📨 AI for Social Impact Deep Dive: Vibe Coding

✍🏼 A Note From the Editor

In February 2025, Andrej Karpathy (formerly of OpenAI and Tesla), coined the term "vibe coding." One year later, it was named the 2025 Word of the Year, entered Merriam-Webster as a "slang & trending" expression, and reshaped how software gets built.

So, what is it and how can you do it? Read on.

💻 A Crash Course on Software Development

Building software used to require lots of humans and lots of capital. A human developer would write code line by line in a programming language (Python, JavaScript, etc.), test it for bugs, debug it, review it with teammates, and eventually push it to production. A new feature could take days, weeks, or months to ship, and entire teams of specialists, including frontend developers, backend developers, database engineers, security reviewers, QA testers, would have to coordinate on every meaningful change.

🤖 So, What Is Vibe Coding?

Simply put, vibe coding is the ability to build software by describing what you want in plain English instead of writing code yourself. You tell an AI tool what you're trying to build, it writes and runs the code for you, and you keep refining until you arrive at your desired outcome.

🛠️ Vibe Coding Tools

Claude Code (Anthropic) The one that started it all. Claude Code is the tool most often credited with kicking off the modern agentic coding moment. It began as an internal experiment by Anthropic engineer Boris Cherny in late 2024 and was released publicly in early 2025. Anthropic now reports that the majority of code at the company is written by Claude Code itself.

Codex (OpenAI) Codex launched in April 2025 as OpenAI's answer to Claude Code. Like Claude Code, it's a terminal-based agent that reads your repo, edits files, runs tests, and commits code. It is one of the fastest-growing developer tools in history.

Antigravity (Google) Launched in November 2025 alongside Gemini 3, Antigravity is where AI agents can plan, code, test in a built-in browser, and even produce screen recordings to prove their work. Its "Manager" view lets you spawn multiple agents working in parallel across different tasks (most other tools do this now, too).

Lovable Where Claude Code, Codex, and Antigravity assume some technical skills, Lovable is built for non-coders. You describe the app you want in plain English in a chat window, and Lovable builds a complete, deployed full-stack web application—frontend, backend, database, the works—without you ever seeing a line of code.

Replit Replit predates the vibe coding moment but has fully reinvented around it. Its AI Agent handles full-stack development including server-side code, databases, APIs, authentication, and deployment—all inside a browser. You don't have to install anything.

🌊 How Do I Vibe Code?

Step 1: Develop the idea. Start with a problem you want solved or something you want to build. Maybe it's a simple internal dashboard that shows your program data at a glance, or a tool that converts board meeting notes into action items.

Step 2: Describe what you want, in plain English. Open your tool of choice. Start with your prompt: "I want to build a simple web app where staff can paste in board meeting notes and get back a list of action items, organized by owner and due date.” If you’re using Lovable or Replit, that’s enough to start with.

If you’re using Claude Code or Codex, the process is a bit more technical. You’ll first need to create a product requirements document (PRD). Ask Claude or ChatGPT to ask YOU questions, one by one to create your PRD. Then you can use the PRD to start vibe coding. Note, you will need an IDE (like Visual Studio, Windsurf, or Cursor) that have integrations with Claude Code or Codex. When in doubt, ask AI for help.

Step 3: The AI generates the code. Behind the scenes, the AI writes the actual programming code needed to build what you described. In the older, pre-agentic workflow, the AI would output the code into a chat window, and you'd have to copy and paste it into a code editor, and follow tedious instructions to build out the app. But no more!

Step 4: Now, the agent runs the code for you. Modern agentic tools skip the copy-paste step entirely. Claude Code, Codex, Antigravity, Lovable, and Replit all execute the code themselves—spinning up a working app in seconds, often with a live preview you can click around in.

Step 5: Test the output. Use the app the way a real user would. Does it do what you asked? Does it look right? Are there obvious bugs? If you're working with any kind of data, pay attention to whether the app handles it cleanly.

Step 6: Iterate. If something's off, you tell the AI in plain English: "The action items are coming out as one big paragraph. Can you format them as a clean checklist instead?" The AI updates the code, re-runs the app, and shows you the new version. Round and round until it matches what's in your head.

Step 7: Deploy (or don't). Totally up to you!

👋🏼 About AI for Social Impact

I’m Joanna, and I’m on a mission to help folks in the social impact sector understand, experiment with, and responsibly adopt AI. We don’t have time to waste, but we also can’t get left behind.

Let’s move the sector forward together. 💫

♥️ Spread the Love

Spread the love and forward this newsletter to anyone who might benefit from a dose of AI inspo!