Descriptions:
Ross Mike joins Greg Isenberg for a structured crash course on Claude Code, aimed at beginners and non-technical builders who have been intimidated by the terminal-based AI coding agent. Rather than walking through a single build, the episode focuses on the mental models and workflow habits that determine whether Claude Code produces quality output or expensive noise.
The central lesson is planning. Ross argues that the model is now so capable that poor outputs are almost always caused by poor inputs — specifically, by describing products at a high level instead of breaking them into discrete features. He demonstrates the ask_user_question tool, which instructs Claude Code to surface ambiguities before writing code rather than making assumptions, and explains how this single habit can prevent hours of rework and wasted API tokens. The episode also covers why Ross advises against using agentic auto-run modes (he compares it to letting a new driver use Tesla’s self-driving feature) and how to structure iterative planning-to-build loops that keep the agent aligned with intent.
The session closes with a tips-and-tricks section covering practical setup, prompting strategies, and how to think about token costs during planning versus execution. For anyone who has run Claude Code and been frustrated by the results, or who wants to start with the right habits from day one, this episode provides a clear, opinionated framework for using the tool effectively — delivered by someone who has refined the approach through repeated real-world use.
📺 Source: Greg Isenberg · Published January 19, 2026
🏷️ Format: Tutorial Demo
