Building Secure Services with AI Agents: Practical Patterns
summary
Greenfield payments backend gets built whole service in two days using AI agents but agents introduce regressions and violated PCI DSS checks during code generation. Rapid churn makes the codebase feel like legacy in weeks. Practical patterns are: write concrete specs and split planning from execution, keep context small by limiting MCPs and skills and maintaining agents.md, enforce deterministic rules with CICD guardrails, and run agents inside a sandbox as a restricted Docker container to boost safe autonomy.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 2:30Greenfield AI agents
- 4:17PCI DSS violation
- 6:50Vibe coding limits
- 7:56Agent workflow planning
- 10:01Context hygiene
- 12:54Secure agent sandboxes
- 15:00Agent autonomy tests
- 17:30Summary
related talks
22mBuilding Secure Services with AI Agents: Patterns From Real-World Engineering
11mBuilding Secure Backend Services with AI Agents
36mHow to Safely Deploy AI Agents that Write and Manage your IaC
30mBuilding Tools with Tools: Modern AI-Driven Development
48mSecuring Agentic AI
14mThe 4 Patterns of AI Native Development
39mBuilding Safe AI Agents
40mIn AI We Trust? Securing the Future, One Agent at a Time