Detection Engineering Dispatch 5 Bitter Pills to Swallow RE: Agentic AI w/ Oliver Rochford
Jun 5, 2025
In this discussion, Oliver Rochford, founder of aunoo.ai and writer of the Curious AI newsletter, reveals essential truths about agentic AI. He highlights the misconceptions surrounding autonomy in AI agents, the danger of 'agent washing,' and the need for critical oversight in security operations. Rochford emphasizes the importance of reliable data and human collaboration, challenging listeners to rethink the hype around AI technologies. The conversation underscores the necessity for clear communication and foundational infrastructure to successfully leverage AI in cybersecurity.
AI Snips
Chapters
Transcript
Episode notes
Agentic AI Requires True Autonomy
- Most tools claiming to be agentic AI are just automated step runners or wrappers without true autonomy.
- Real agentic AI requires novel decision-making beyond rigid workflows that demonstrate actual agency.
Prioritize Data Quality First
- Avoid building agents in the SOC atop poor-quality, noisy, or outdated detection signals.
- Focus first on maturing data hygiene and detection quality before deploying agentic workflows.
Risk Shapes Agent Autonomy
- Fully agentic workflows depend heavily on the organization's risk appetite and leadership accountability.
- High error rates make autonomous deployments risky and raise compliance and negligence concerns.