CNV - What Makes Agents Different?

Описание: In this video, Morgan Foster, co-chair of the AI Gateway Working Group and emerging technology expert at Red Hat, explores the conceptual frameworks needed to build a robust infrastructure for the "Internet of Agents". Moving beyond the hype of individual protocols and implementation details, Foster offers a provocative perspective on why managing AI agents is fundamentally an architectural challenge that we’ve already solved in other domains.

00:09 - Introduction by Morgan Foster.
02:24 - The "Magical Faucet" Analogy: LLMs and agents are compared to a stream of "liquid gold"—contextually relevant text—that requires a specific set of "buckets" (infrastructure) to capture and turn into economic utility.
04:48 - The Problem of the Unreliable Executive: Agents are powerful but lack accountability, individual identity, and predictability. Foster discusses how to coax useful work out of these executives without risking your entire database.
07:08 - Lessons from Mechanical Turk: By examining Amazon's Mechanical Turk from 2005, the video reveals how long-standing patterns like redundancy, quorums, gold standards, and reputation systems are being adapted for modern AI agent evaluation frameworks like MLflow.
12:03 - Sandboxing and Security: Drawing parallels to VPS (Virtual Private Server) hosting, Foster explains why extensive isolation and sandboxing at every layer—including using tools like Pydantic's Monty—are critical for running untrusted agent code safely within your infrastructure.
21:00 - Q&A

Learn more - github.com/cloudnativevan/meetup/issues/20

#aiagents #artificialintelligence #infrastructure #llm #redhat #kubernetes #techtalks #machinelearning #cybersecurity #softwarearchitecture