Summary
Developer and creator Theo argues that most engineers are bad at prompting, not because they under-engineer their prompts but because they over-engineer them. In this video he makes the case that prompt engineering hype missed the point, and that the real unlock is stepping out of the engineering mindset and asking agents to do far more than most people currently expect.
He frames the discussion around a well-known XKCD about the gap between how normal users and engineers think about problems, then walks through practical examples in T3 Code. These include asking one agent thread to spin up two Opus 5.5 delegates to explore performance improvements, then start a new thread with Grok 4.7 to implement the findings, as well as generating a context window visualization simply by asking for it. He also explains when repeated instructions should be turned into skills or CLAUDE.md files.
The second half returns to engineering with this mindset. Theo demonstrates the T3 CLI triage command, a built-in tool that helps debug local problems even when the desktop app will not open, and explains how a simple, prompt-driven approach shaped it. The video also includes a sponsor segment for the Macroscope code review tool.
📺 Source: Theo – t3․gg · Published October 18, 2026
🏷️ Format: Opinion Editorial
