Benchmarking AI infrastructure

Kai Hendry and Vincent De Smet try Rex, test whether Claude Code and Codex can message each other, and discuss API keys, zero trust, agent identity and a wild week of security exploits.

Kai Hendry and Vincent De Smet debate build versus buy in the age of AI, infrastructure testing, agent workflows and the human bottlenecks in enterprise delivery.

Conference plans, agent harnesses, AI-assisted Terraform reviews, and the engineering practices that make delivery faster.

Always-on agents promise freedom, but reviews, CI, model costs and human accountability still shape what we can build.

AI agents can write code faster. But every change can still get stuck waiting for an approval, a merge button or an hour-long CI run.

Is the terminal becoming the IDE, or is the IDE becoming an agent? Kai has spent two days living in Orca, a terminal-shaped harness that runs Claude Code, Codex and friends side by side, with worktrees, a diff pager you can comment on, and a mobile app that talks to your laptop over your own LAN.

This episode is a conversational walkthrough of Kiro Crew (an AWS-hosted agent gateway and dashboard), how it uses the agent client protocol (ACP), what running and extending it looks like, and practical lessons about memory, embeddings, deployment and costs.

Quick take: a technical conversation about infrastructure-as-code, emerging frameworks, and using AI agents to automate and benchmark cloud operations.

For vulnerability management https://www.invicti.com/ previously known as https://kondukto.io/ was what I was referring to!

https://github.com/kaihendry/members.cardinhamsports.org/commits/main/ is a demonstration of using https://github.com/kaihendry/sloc-sensor