The question this episode answers: When an AI agent has write access to your production system and it's wrong — what survives?

The answer: The teams that survive machine-speed engineering are the ones who never gave AI agents access to the original — only to a tested copy they can walk away from in under sixty seconds.

What you'll learn:

• Why 'capability without judgment' is an infrastructure risk category, not just a philosophical observation
• The three specific failure modes AI agents create in production: schema events, data modification, and secret leakage
• Why the blast radius of an AI mistake is velocity-shaped, not magnitude-shaped
• What zero-trust applied to non-human actors actually requires architecturally
• The three tests an isolated AI-agent environment must pass to be genuinely useful
• How recovery speed determines whether containment is sustainable as a daily practice — not just a policy on paper

References:

• John Lennox interview, 'Does This Prove That God Is Real?', Diary of a CEO with Steven Bartlett (June 2026): https://www.youtube.com/watch?v=dBb_Xxj37VQ
• John Lennox, '2084: Artificial Intelligence and the Future of Humanity', Zondervan (2020): https://www.zondervan.com/9780310109761/2084/
• Jesus Christ — first-century historical figure referenced in Lennox's personal testimony on forgiveness: https://en.wikipedia.org/wiki/Jesus

About this episode: Built for the CTO, CISO, or founder running AI agents in production-adjacent environments who hasn't yet done a formal audit of what those agents can touch and modify. The AI-agent containment assessment is at rediacc.com/ai-safety.

About this show: Discontinuity is a weekly briefing on infrastructure resilience, published by Rediacc. The narrator is an AI library voice rendering research the team writes. Sponsor reads are for Rediacc's own products.