The InfoQ Podcast

Agentic Systems Without Chaos: Early Operating Models for Autonomous Agents

64 snips
Mar 25, 2026
Joseph Stein, an experienced software architect who builds centralized AI platforms, discusses running autonomous agents without chaos. He outlines how agentic systems differ from traditional automation. Short takes cover boundaries and orchestration, new security risks like prompt injection, observability and explainability needs, platform design, cost tradeoffs, and the surprises that come with rapid adoption.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
ANECDOTE

Open Source Agents Spark Rapid Experimentation

  • Joseph Stein recounted following OpenClaw and how open-source agents run on Raspberry Pi, sparking rapid experimentation.
  • He noted OpenClaw's author later joined OpenAI and that VCs immediately cloned hosted sandboxes.
ADVICE

Instrument AI With Production Observability

  • Build AI observability similar to system observability: track prompts, tool calls, orchestrations, and prompt history for SDLC and auditing.
  • Stein recommended capturing prompt/tool flows so teams can SDLC agent behaviors and provide explainability.
INSIGHT

Agentic Systems Create New Security Risks

  • New attack surfaces include prompt injection, tool misuse, token-denial attacks, and poisoned model outputs that can insert backdoors into generated code.
  • Stein referenced Bruce Schneier's Morris worm analogy and papers showing LLMs can be trained to output malicious code.
Get the Snipd Podcast app to discover more snips from this episode
Get the app