Friction in agentic coding is engineering judgment

A company that marketed ‘ship without friction’ shipped a security incident via a configuration change. The pushback you feel when reviewing an agent’s output is your expertise (see: comprehension debt compounds silently) at work, not inefficiency to optimize away. A senior engineer’s instinct to pause (see: supervised agents outperform fully autonomous ones), question an assumption, or insist on a test is exactly what prevents incidents. Removing that friction removes the safety net.

Cristina Poncela Cubeiro describes herself as a ‘native AI engineer,’ someone who learned to code with these tools rather than before them. This perspective reveals the other side: if your judgment was never formed by years of friction, you may not feel it when you should. The instinct to pause and question an assumption is a skill built through thousands of hours of debugging, reviewing, and making mistakes. Removing all friction from the workflow accelerates output but starves the formation of that instinct in the next generation of engineers.