『Ep5: John Capobianco — A network automation builder who turned early ChatGPT experiments into a production-grade ReAct agent now weighs how much human oversight and orchestration remain essential once agents handle intent, verification, and collaboration in regulated environments.』のカバーアート

Ep5: John Capobianco — A network automation builder who turned early ChatGPT experiments into a production-grade ReAct agent now weighs how much human oversight and orchestration remain essential once agents handle intent, verification, and collaboration in regulated environments.

Ep5: John Capobianco — A network automation builder who turned early ChatGPT experiments into a production-grade ReAct agent now weighs how much human oversight and orchestration remain essential once agents handle intent, verification, and collaboration in regulated environments.

無料で聴く

ポッドキャストの詳細を見る
John Capobianco revealed that his production ReAct agent can now parse high-level intent, verify proposed changes against live network state, and coordinate with peer agents before any human sees the plan, yet he still routes every change through mandatory approval gates in regulated environments. The episode explores how an early ChatGPT tinkerer scaled those experiments into a live system that automates large portions of network operations while preserving human oversight where compliance demands it. • Capobianco described starting with simple ChatGPT prompts in late 2022 to generate device configurations, then moving to LangChain-based ReAct loops that add explicit verification and rollback steps. • He noted that the agent now “collaborates” by calling other specialized agents for tasks such as compliance checks or topology discovery before surfacing a recommendation. • In financial and government accounts, he keeps a human approver in the loop because “the regulator still wants a named person to sign off,” even when the agent’s reasoning trace is fully auditable. • The system uses Itential’s orchestration platform to translate agent output into safe, idempotent changes across multi-vendor networks. • Capobianco runs side experiments on automateyournetwork.ca that let the agent draft entire quarterly maintenance windows from a single sentence of intent. • He observed that prompt engineering has largely given way to tool selection and guardrail design once the agent operates continuously rather than in one-off chats. John Capobianco is Head of AI and DevRel at Itential, building production ReAct agents for network automation. Find John Capobianco: automateyournetwork.ca: https://automateyournetwork.ca YouTube: https://www.youtube.com/@johncapobianco2527 X: https://x.com/John_Capobianco LinkedIn: https://www.linkedin.com/in/john-capobianco-644a1515
adbl_web_anon_alc_button_suppression_t1
まだレビューはありません