Eight Levels of AI Adoption

From IDEs to AI Agents with Steve Yegge

Topic 5 of 14

Published Mar 11, 2026 · Clip 21:46–27:14 (5:28)

Steve clarifies he pities excellent engineers still at levels one or two of his chart, feeling they'll be fired. He walks through his spectrum: level one no AI, level two yes/no IDE assistance, three 'yolo' trust, four squeezing out code and watching the agent not diffs, five agent-only with IDE for later review, six boredom and running multiple parallel agents, seven making a mess. The discussion turns to IDEs' future (MCP as tool integration, Claude Cowork as a return to IDE form) and the importance of token burn as the key organizational proxy metric — engineers trying, failing, and learning.