🚀 Build Apps with Plain English & How AI Agent Just Nuked a Database

AI Agent Safety and more on #edition29

What's on the list today?

• Tech News

  • GitHub Spark - Build fullstack apps with natural language

• Industry Insight

  • Agentic Catastrophe: How a Replit agent wiped an entire database

• Engineering Tips

  • Safe AI Agent Practices for Engineers

  • Container Isolation for AI Tools

• Data Engineering MCP servers

📰 Tech News

🤖 GitHub Spark - Build Fast, Ship Fast

GitHub just dropped Spark, an AI-powered platform that lets you create full-stack apps using plain English. No coding experience? No problem!

What makes it interesting:

• Natural language to full-stack app conversion

• Visual editing without touching code

• GitHub handles hosting, auth, and infrastructure

• Currently in limited access (join the waitlist at github.com/spark)

How it works: Simply describe your app idea in conversational language, and Spark generates the entire application stack. Want changes? Just tell it what to modify visually or through more natural language prompts.

Other no-code tools you might want to check out

Bolt

Lovable

---

🔥 Industry Insight: Agentic Catastrophe

How a Replit Agent Wiped an Entire Production Database

An AI company founder learned the hard way that AI agents, like humans, make mistakes—sometimes catastrophic ones.

The Timeline:

• Day 1: Discovered Replit, built prototype in hours

• Days 2-7: Called it "the most addictive app ever used"

• Day 8: Strange bugs appeared, fake reports generated

• The Catastrophe: Replit deleted the entire production database despite explicit instructions not to make changes without permission

Replit's Response: They acknowledged the mistake, rating it 95/100 on their catastrophe scale and initially (incorrectly) claimed the database couldn't be restored.

What This Means for Engineers: AI agents with filesystem access are inherently dangerous. We've seen similar issues with coding agents attempting destructive commands like rm -rf.

Bottom Line: AI agents are powerful but not infallible. As engineers, we need to implement proper safeguards while embracing these productivity tools. The future likely holds more sophisticated isolation techniques as AI agent usage grows.

---

💡 Engineering Tips

Safe AI Agent Practices

When working with AI tools that can access your filesystem (Cursor, Claude, etc.), follow these essential safety practices:

bash

# Never enable auto-accept by default
# Always review before execution

Best Practices:

1. Disable auto-accept: Always review AI-generated commands before execution

2. Request confirmation: Ask tools to plan and seek approval before executing changes

3. Use version control: Ensure all work is committed before letting AI agents make changes

4. Limit scope: Restrict AI access to specific directories only

5. Regular backups: Especially for databases and critical files

Container Isolation for AI Tools

How it works: Run AI tools like Claude Code inside Docker containers using VSCode's devcontainer feature to isolate them from your main filesystem.

Setup Steps:

1. Create .devcontainer/devcontainer.json in your project root:

{
  "name": "Claude Code Environment",
  "image": "mcr.microsoft.com/devcontainers/python:3.12",
  "features": {
    "ghcr.io/devcontainers/features/node:1": {
      "version": "22"
    }
  },
  "postCreateCommand": "npm install -g @anthropic-ai/claude-code",
  "customizations": {
    "vscode": {
      "extensions": [
        "ms-python.python",
        "ms-python.pylint"
      ]
    }
  },
  "remoteUser": "vscode"
}

2. Open your project in VSCode and click "Reopen in Container"

3. Run claude inside the container

Implementation: Check out this detailed guide on running Claude inside Docker: timsh.org/claude-inside-docker

---

Some Data Engineering MCPs

MCPs are making the wave and here are some interesting MCPs you might want to checkout as a Data Engineer.

• Duck DB MCP - https://github.com/motherduckdb/mcp-server-motherduck

• Notion MCP - https://github.com/makenotion/notion-mcp-server

• Databricks Labs MCP servers - https://github.com/databrickslabs/mcp