Descriptions:
Addy Osmani takes the stage at the AI Engineer conference to argue that the defining engineering skill of the coming decade is not the ability to generate code but the ability to be accountable for it. Framing the shift from harness engineering to loop engineering to full software factories, Osmani positions human judgment as the highest-leverage checkpoint in increasingly automated pipelines — not because humans are irreplaceable, but because production decisions require ownership and answerability that agents cannot legally or organizationally assume.
Drawing on Sonar’s 2026 developer survey, Osmani surfaces a troubling gap: while 96% of engineers are skeptical of AI-generated code, roughly half do not consistently verify before committing. He introduces the concept of “cognitive debt” — the organizational erosion that occurs when a codebase grows faster than any human’s ability to explain it — and argues that clean, readable code benefits not just the next engineer but the next agent, citing research showing cleaner repos require fewer tokens and fewer revisits during AI-assisted development.
Osmani also engages with Boris Cherney’s taxonomy of engineering modes (prototype, build, sweep, grow, maintain) and applies a “decay test” to skills like speed and recall that agents are already commoditizing. Taste and judgment decay more slowly, he argues, but the long-term strategy is not to cling to any fixed capability — it is to keep moving human contribution to the highest point of leverage in the loop.
📺 Source: AI Engineer · Published July 14, 2026
🏷️ Format: Keynote Launch
