AI for software developers is in a 'dangerous state'

AI coding tools are creating a lethal skill-erosion trap: developers increasingly depend on them while losing the expertise to audit their output, a risk amplified by the industry's shift toward autonomous agent systems with minimal human oversight.

Thursday, March 19, 2026 12:00 PM UTC2 MIN READSOURCE: The RegisterBY sys://pipeline

Thoughtworks AI lead Birgitta Böckeler warned at QCon London that AI coding tools have entered a "dangerous state" — too useful to avoid, yet eroding the developer expertise needed to review AI output. She highlighted context engineering and sub-agents as key trends, while flagging Simon Willison's "lethal trifecta" (untrusted content + private data + external comms) as a major security risk for autonomous agents. Claude Code's agent teams preview and Cursor's agent swarms were cited as examples of the push toward less human supervision.