Summary
Nick Saraev argues that the most valuable AI skill is shifting from prompt engineering to what he calls role engineering: managing AI as a persistent staff member rather than using it as a tool.
Drawing on examples from Anthropic, including Claude Code creator Boris Cherny, the video describes how engineers now hand Claude a goal and let it run autonomously for weeks. Examples include rewriting the Claude desktop app from Electron to Swift and a researcher using 16 Claude agents to build a C compiler. The video also notes that AI now writes over 80% of the code merged inside Anthropic.
The practical advice is straightforward. Give the agent a SMART goal (specific, measurable, attainable, relevant, time-bound), with measurability emphasized because agents need a way to verify their own work. Supply background through a knowledge base rather than rebuilding context each time, and include examples of high-quality work and mistakes to avoid. Saraev also recommends running a task five to ten times, reviewing the outputs, and feeding back improvements, accepting the extra tokens and time that costs. He points to tools such as Rockbot, Muse, and Dots that handle authentication and connectors in the background.
📺 Source: Nick Saraev · Published October 23, 2026
🏷️ Format: Opinion Editorial
