Concrete, achievable projects that build confidence — from portfolio site to AI agent
Turn what you learned into a concrete stack decision.
01AI Agent ToolboxBuild autonomous AI agents that can read, write, and actWant the shortlist in your inbox?
Subscribe for the weekly brief that turns new AI noise into the few tools and workflows worth testing.
Curated bundles that help you move from this guide into a working stack.
The essential toolkit for building AI agents with the Model Context Protocol. This pack includes MCP servers, agent frameworks, and workflow orchestration tools that let AI agents interact with real systems from GitHub to databases to Slack.
Open pack →A curated collection of essential tools for DeFi development. From smart contract frameworks to Web3 libraries, this pack covers the full stack for building decentralized finance applications on Ethereum and EVM chains.
Open pack →A complete starter kit for Solana development. This pack combines Web3 libraries, AI agent SDKs for Solana, trading bots, and learning resources to get you from zero to mainnet deployment fast.
Guide
Installing OpenClaw: Zero to Running Agent
Get your own AI agent with personality, memory, and Telegram access in under 30 minutes
Guide
Keeping Your Stuff Safe: Security for AI Builders
What to protect, what's safe to share, and the mistakes that cost people real money
Guide
Ship Your First Claude Code Skill in 30 Minutes
Turn any workflow into a reusable slash command — no framework needed
The goal of your first project isn't to build the next unicorn. It's to prove to yourself that you can make a computer do something useful with AI help. That confidence compounds — each project teaches you patterns you'll reuse forever.
These 5 projects are ordered from easiest to most ambitious. Each one teaches a fundamental skill.
Time: 30-60 minutes | Teaches: HTML/CSS, deployment, working with AI tools
A single-page site that says who you are and what you're building. It's the most common first project for a reason — you end up with something real you can show people.
npm install -g vercel # Install Vercel CLI
vercel # Deploy (follow the prompts)
Time: 1-2 hours | Teaches: APIs, bot tokens, environment variables, server processes
A Telegram bot that responds to commands. This is foundational — bots are one of the most common things you'll build with AI.
/start, /help, and a custom commandCreate a bot: Open Telegram, search for @BotFather, send /newbot, follow the prompts. You'll get a token like 123456:ABC-DEF1234ghIkl-zyx57W2v1u123ew11.
Set up your project:
mkdir my-telegram-bot && cd my-telegram-bot
bun init -y
bun add grammy
Tell your AI tool: "Create a Telegram bot using grammy that responds to /start with a welcome message, /help with a list of commands, and /quote with a random motivational quote. Use the bot token from a BOT_TOKEN environment variable."
Create your .env:
BOT_TOKEN=your-token-from-botfather
bun run index.ts
/start.Time: 1-2 hours | Teaches: REST APIs, JSON parsing, scheduling, conditional logic
A script that checks cryptocurrency prices and alerts you when they cross a threshold. This is a great introduction to working with external APIs.
mkdir price-alerts && cd price-alerts
bun init -y
Tell your AI tool: "Create a price alert script that: (1) fetches Bitcoin and Ethereum prices from the CoinGecko API (free, no key needed), (2) compares against configurable price targets, (3) sends a Telegram message via bot API when a target is crossed. Store targets in a JSON config file. Use Bun with fetch API."
The CoinGecko API is free and needs no key:
https://api.coingecko.com/api/v3/simple/price?ids=bitcoin,ethereum&vs_currencies=usd
# Run every 5 minutes with cron
*/5 * * * * cd /path/to/price-alerts && bun run index.ts
Time: 2-3 hours | Teaches: MCP protocol, tool definitions, AI tool integration
MCP (Model Context Protocol) is how AI tools discover and use external capabilities. Building a simple MCP server teaches you the protocol that powers AI agent ecosystems.
search_notes and read_notemkdir my-notes-mcp && cd my-notes-mcp
bun init -y
bun add @modelcontextprotocol/sdk zod
Tell your AI tool: "Create an MCP server using @modelcontextprotocol/sdk that exposes two tools: (1) search_notes — takes a query string and returns matching filenames from a ~/notes/ directory by searching file contents, (2) read_note — takes a filename and returns the full contents. Use stdio transport. Make it work with markdown files."
Create some test notes:
mkdir -p ~/notes
echo "# Meeting Notes\nDiscussed the new API design" > ~/notes/meeting-2026-02-25.md
echo "# Ideas\nBuild a price alert bot" > ~/notes/ideas.md
{
"mcpServers": {
"my-notes": {
"command": "bun",
"args": ["run", "/path/to/my-notes-mcp/index.ts"]
}
}
}
Time: 2-3 hours | Teaches: LLM API calls, prompt engineering, chaining operations
A command-line tool that takes a URL, fetches the page content, and produces a summary using an AI model. This teaches you the most fundamental pattern in AI engineering: fetch → process → generate.
mkdir summarizer && cd summarizer
bun init -y
bun add @anthropic-ai/sdk
Tell your AI tool: "Create a CLI script that: (1) takes a URL as a command-line argument, (2) fetches the page HTML, (3) extracts the main text content (strip tags, scripts, styles), (4) sends the text to Claude API with a prompt asking for a structured summary with key points, (5) outputs the summary to stdout. Use the Anthropic SDK. Read API key from ANTHROPIC_API_KEY env var."
Set up your API key:
echo "ANTHROPIC_API_KEY=sk-ant-..." > .env
bun run summarize.ts https://example.com/interesting-article
After these 5 projects, you've learned the fundamental patterns:
Now look at your own life for problems to solve. The best projects come from personal pain points:
The jump from "tutorial project" to "real project" is smaller than you think. You already have the patterns — now apply them to problems you actually care about.
Want to level up further? Learn about OpenClaw — an AI agent framework that gives your AI a personality, memory, and autonomy.