QCon London: AI for developers is in a 'dangerous state'

At QCon London, Thoughtworks AI lead Birgitta Böckeler warned that AI for developers is in a 'dangerous state': it's too useful to ignore, yet using it erodes the experience developers need to review its output. Key themes include context engineering, sub-agents and agent swarms reducing human oversight, security risks like prompt injection and the 'lethal trifecta' (untrusted content + private data access + external communication), rising costs from agentic workflows, and the paradox that developers are losing hands-on skills precisely because AI is doing the work for them. OpenAI's concept of 'harness engineering' is mentioned as a potential path toward safer agent environments.