AI 2026-10-27 3 videos

Building Fast, Measurable, Governed AI Agents

AI agents need tight identity, permissions, and monitoring before they can safely scale.

Summary

AI operations need infrastructure that moves quickly while measuring models, tool use, spending, and outcomes. Optiver focuses on same-day deployment through an on-premises, modular platform, while Runlayer focuses on visibility that exposes prompt injection, excessive permissions, session drift, and runaway costs across more than 300 agents. Governance adds a workforce model: every agent needs a unique identity, a team, a human sponsor, a defined role, bounded authority, and escalating oversight. The shared direction is clear, but the approaches differ between shipping models efficiently, detecting operational risk, and assigning accountability for every action.

The videos

How a Trading Firm Ships New AI Models the Same Day - Noyan Tokgozoglu, Optiver ↗ AI Engineer

Optiver's on-premises AI platform supports 10 million-plus daily trades across 10-plus locations and can benchmark and ship a new frontier model the same day while tracking cost, identity, tools, and performance.

300 Agents, 60 People: How We Secured Our Own AI Agents - Aiden Sochowski, Runlayer ↗ AI Engineer

Runlayer, a 50–60-person company running more than 300 agents daily, uses a central gateway and trained classifiers to monitor every session and tool call for attacks, excessive permissions, task drift, and cost spikes.

Why Are We Still Governing AI Agents Like Software? - Tim Tyler, Gravitee ↗ AI Engineer

Tim Tyler's Agent Accountability framework treats agents as a digital workforce through Assignment, Alignment, and Administration, giving each one an identity, team, human sponsor, bounded authority, and escalating oversight.