Summary
Craig Hewitt, founder of podcast platform Castos, demonstrates how he built a personalized AI morning briefing delivered as a private podcast using three tools: Grokbot as the AI agent orchestrator, Fish.audio for generative text-to-speech, and Castos for podcast distribution. The entire pipeline was configured through conversational prompting in approximately 20 minutes — no code written manually.
The workflow begins with Craig describing the briefing he wants in Grokbot: a daily digest of his calendar, email, and relevant industry news, delivered at 5:00 a.m. Grokbot autonomously discovers Fish.audio’s MCP server, handles OAuth authentication, and generates voice samples. After rejecting the initial synthetic voice, Craig requests a custom female British voice, which Grokbot designs directly inside the Fish.audio browser dashboard. The final audio briefing is pushed via the Castos REST API to a private RSS feed accessible in any podcast app.
The video highlights two practical agentic building patterns: first, opening a session by asking the agent to “analyze and ask me relevant questions” before building — a clarifying question loop that surfaces ambiguities early. Second, it demonstrates real MCP-based tool chaining across calendar data, web news sources, audio generation, and podcast publishing without manual integration code. Craig also shows an honest limitation: X’s developer API costs led him to drop Twitter as a news source and fall back to standard web search, illustrating the kind of real-world constraint navigation that distinguishes hands-on builds from scripted demos.
📺 Source: Craig Hewitt · Published September 03, 2026
🏷️ Format: Tutorial Demo
