We are building data breach machines and nobody cares
Summary
The post uses Castlevania metaphors to frame AI agents as Dracula-like threats within security contexts and argues that the security challenge lies in the architecture and orchestration of agentic workloads rather than purely technical AI capability. It covers core concepts such as planning, ReAct, memory vs context, multi-agent orchestration, and the fragmented state of industry standards, ultimately advocating for traditional, robust security controls and a security-by-design mindset to mitigate risks.