Container Solutions TV Should Autonomous AI Agents Run Your Business?
13 snips
Feb 19, 2026 
Adrian Cockcroft, cloud computing pioneer now exploring generative AI and agentic systems. He discusses developer-on-demand workflows and how agentic coding speeds prototyping. Short takes cover accountability, security, observability, read-only shadow rollouts, and how management and platform skills will shape agent-driven work.
AI Snips
Chapters
Books
Transcript
Episode notes
Run Agents With Human Oversight
- Use conditional autonomy for agents and keep humans accountable at design time.
- Allow agents limited actions, require human approval for high-risk tasks like payments, and build clear ownership and compliance trails (Önsel Akin and Chris Vermeulen discussion).
Agents Are Opaque Acting Workers
- Agents differ from batch jobs because they can read, write, call APIs, discover endpoints and act across domains.
- That operational complexity creates opaque internal decisions and shifts visibility needs from deterministic logs to observability for goals and actions (Önsel Akin explanation).
Red Team Agents Like Security Systems
- Red team agents aggressively for prompt injection, sleeper backdoors and non-deterministic failures.
- Try adversarial emails, malformed inputs and switching model safety levels to expose behaviors that only appear under production conditions (Önsel Akin examples).
