How to Build a Full SaaS App with AI Agents While You Sleep

Watch Episode

Listen to Episode

Show Notes

Nathan, a forward deployed AI engineer at Google Cloud, joins Charles Brun to explain how he vibe-coded an open-source Slack clone, Slawk, in just two weeks using Claude Code, without touching a line of code. His viral LinkedIn post sparked both excitement and skepticism, and in this episode he breaks down exactly how he did it. He started by setting up the backend and frontend with Claude using Node, Postgres, and React, then used Claude’s browser tooling to iteratively match Slack’s UI. The core workflow was three agents in parallel: a QA agent that constantly tested the app and filed issues with screenshots and GIFs, a dev agent that fixed issues and wrote Playwright tests, and a CTO agent that orchestrated the whole process. He explains why Claude Code’s subagents were the key unlock, enabling parallelism and reducing context overload. The conversation also tackles the biggest objections around scalability, security, maintenance, and integrations, and closes with a broader thesis: AI-native software can be built faster and differently than incumbent SaaS products built for a pre-agent era. Claude Code supports custom subagents, which Anthropic documents as a way to delegate specialized tasks.

Key Takeaways

• Vibe coding is enabling non-technical and semi-technical builders to ship production software
• Claude Code subagents enable true parallelism and prevent context rot
• Three agent architecture (QA
• dev
• CTO) can maintain a codebase autonomously
• Cloud managed services absorb most scalability and security concerns for vibe coded apps
• The Slack of tomorrow has one integration (an MCP server) instead of 400
• Finding bugs is now a bigger bottleneck than fixing them
• Incumbent SaaS tools can't easily rebuild themselves from scratch the way vibe coders can
• Total cost of a full Slack clone vibe coded in 2 weeks was about 200 dollars
• QA as a service is an emerging category driven by vibe coding
• Software engineer identity pushback is part of why AI adoption meets resistance

Guest

Nathan Cavaglione — AI Engineer, Google Cloud (Forward Deployed Engineer)

LinkedIn profile