Hey makers 👋

Another week of AI tools that could change how you build and ship. This week:

• Google's free terminal AI tool

• Claude's astounding 500M app milestone

• The prompting techniques powering YC's best startups

Let's dig in!

🚀 Ship Faster

Share AI-powered apps with Claude

Anthropic just turned Claude into your personal app factory. Type what you want, and it'll write the code, host it, and make it shareable. Over 500 million apps have been created since launch, last August → 🙀

Free users can build and share, while Pro users ($20/month) get advanced features. No API keys, no hosting headaches, no backend management. Get started.

Why it matters: Got a SaaS idea but hate dealing with infrastructure? Try starting here first. Validate your concept with actual working prototypes before you invest in proper development.

Google's Free CLI Tool: Vibe Code From The Terminal

Google just fired a major shot in the AI developer wars by releasing Gemini CLI - a free, open-source terminal agent that brings their best coding model (Gemini 2.5 Pro) directly to your command line. This isn't just another coding assistant; it's Google's play to steal developers from OpenAI and Anthropic by removing the biggest barrier: cost.

Why it’s awesome

• Generous Free Tier: 60 requests per minute and 1,000 daily queries at zero cost.

• Currently the Best Coding Model: Gemini 2.5 Pro is the highest-ranked model on WebDev Arena right now, with a massive 1M token context window = it can understand and work with your entire codebase at once.

• Beyond Just Code: While it excels at development, you also get access to Google's image (Imagen) and video (Veo) generation models - all from your terminal.

This is "vibe coding from the command line" - you stay in flow state without context switching between browser tabs, design tools, and IDEs.

Getting Started

npm install -g @google/gemini-cli
gemini

The tool reads GEMINI.md files for project-specific instructions, supports Model Context Protocol (MCP) for extensions, and works on Windows, Mac, and Linux.

Why This Matters

Google is weaponizing free access to shift entire developer workflows into their ecosystem. But hey! This means access to cutting-edge AI dev tools without the API bills that usually eat into bootstrapped budgets.

Get Started: GitHub Repository | Google AI Studio

🧠 Level Up Your Game

Y Combinator Just Revealed How Top AI Startups Are Prompting

Y Combinator pulled back the curtain on what's happening inside their best AI startups, and the biggest revelation isn't about the prompts themselves—it's about meta-prompting, and it is a technique that's helping founders ship AI features faster than ever.

Hmmm, buuuuut …

What Is Meta-Prompting?

Meta-prompting is using AI to improve your own AI prompts. Instead of manually tweaking prompts through trial and error, you feed your existing prompt plus examples where it failed into an LLM and ask it to make it better.

How to Meta-Prompt

Step 1: Document Failures 

Save examples where your current prompt produces poor results:

Original prompt: "Write marketing copy for this feature"
Bad output: Generic, boring copy that doesn't convert

Step 2: Ask AI to Improve

"I have this prompt: [your prompt]
But it's producing outputs like: [bad examples]
I need it to: [specific goals]
Please rewrite this prompt to be more effective."

Step 3: Test and Iterate 

You can also have different models improving each other’s prompts.

Why It Matters

As one YC partner put it, meta-prompting feels like "learning how to manage a person" - you're teaching AI to communicate what it needs to make good decisions. While it is a pain, at times, it can significantly accelerate your dev cycle. Dig in: check this Reddit’s thread on Meta Prompts

🌶️ The Funny Slice

🎧 Heard on X…

Keep shipping! 🚀

P.S. - Got an AI tool that's moving the needle for your business? Building an AI tool that helps fellow hackers? Hit reply and tell me about it. Always looking for real-world wins from our community.