Summary
As AI agents mature from question-answerers into operators of real business software, the practical question becomes: how do you actually hand off a large, multi-step project to an agent? Brian Casel addresses this directly with a hands-on walkthrough of two projects he delegated to Claude Code — a full email marketing overhaul and bulk transcript retrieval for over a hundred training videos.
The video introduces a useful mental model: every app your agent needs to interact with has three possible doors — MCP (purpose-built for agents), API (developer-facing but increasingly agent-accessible), and the browser (the fallback for when the other two are unavailable or limited). Casel demonstrates all three using Kit (formerly ConvertKit) as the primary example. Kit’s MCP handles metadata and subscriber management, the Kit API handles custom form integrations, and browser automation via the Vercel agent-browser skill fills the gap for visual workflow editors that neither MCP nor API can reach.
A particularly concrete comparison emerges when Casel tests the Vercel agent-browser skill against Claude’s native browser extension (Claude in Chrome) for the same task — finding agent-browser marginally faster and more reliable. For anyone looking to move beyond one-off prompts into sustained, multi-tool agentic work, this video offers a reusable framework and a realistic look at where the seams between these integration layers actually fall in practice.
📺 Source: Brian Casel · Published July 28, 2026
🏷️ Format: Tutorial Demo
