Summary
Robert Blumoff described Akamai’s move into GPU-backed cloud inference and argued that production AI demands scale, reliability, and tighter controls than lab deployments. His core recommendation was “least capability”: give enterprise agents only the tools required for their task, using the agentic harness as the control point for visibility and fault tolerance.
Key Topics
5 key topics from Robert Blumofe and Matt Rosoff at HumanX Amsterdam 2026. Thicker branches were mentioned more often in the talk.
Trusting AI in the Wild
Click a topic to open it
Key Takeaways
- Apply least capability when designing enterprise AI agents.
- Remove code execution unless the agent truly needs it.
- Use the agentic harness as a visibility control point.
- Plan for reliability failures before scaling to millions of interactions.
- Differentiate personal agents from task-specific enterprise agents.
“Once you get out in the wild, they really do.”
Share