Who picks the model when a Claude Code workflow fans out
Workflow agents can quietly run on your priciest model. The four rungs that pick each agent's model, a probe that proves it, and how to own Claude's guide.
Read the post
Security, GRC & GenAI written from the field.
I'm Chad Norman. I've spent the better part of two decades in IT and cyber security. These days I hunt threats, architect solutions and dig into GenAI and securing autonomous systems. I like figuring out how things work, how they break and how to make them better. This is where I share what I learn, what I build, and the odd rabbit hole.
👋, I'm Chad.
InfoSec · CloudSec · Threat hunter
From threat hunting to cloud security, with a habit of tinkering with GenAI and automation. The opinions here are my own.
Tools
Free online tools I share, starting with Stickerbook, a whiteboard that keeps your boards in your browser.
- Security
- GRC
- GenAI
- LLM security
- Zero trust
- NIST CSF
- Threat modelling
- No tracking
- No cookies
- No pop-ups
- Just RSS
Latest writing
Fresh from the field
The newest posts, in date order.
No AI Act yet, but POPIA, PCI DSS and the Joint Standards already govern your GenAI
South Africa's AI policy stalled, the rules did not. Five controls from POPIA, the Joint Standards and PCI DSS that already apply to GenAI in a regulated firm.
Read the post
Claude Code runs A/B tests on your workflow, and one setting turns them off
Claude Code runs A/B tests from a remote flag service that release notes never mention. How to see yours, switch them off, and why vendors should say so.
Read the post
Just-in-time access shrinks an agent's window, not its reach
Short-lived credentials limit how long an AI agent holds access, not what it can do. Put hard limits in code first, then shorten the tokens.
Read the post
Cloudflare's free plan is where your first project should live
This blog runs on Cloudflare's free plan for R0 a month. What a developer starting out gets there, what they still own, and where the deal stops.
Read the post
Give your AI workflows the least autonomy that works
Autonomy is a permission you grant a model, not a feature it has. Five levels for AI workflows, and how to pick the lowest one each step can live with.
Read the post
Topics
Three things I write about
Notes
Notes from the desk
Short ones: things I learnt, links worth reading, quick takes.
Detections.ai: a shared library of detection rules
A free community library of detection rules with an AI assistant that writes, refines and translates them. Worth a look before you write a rule from scratch.
Your coding agent can hold the lethal trifecta too
Simon Willison's lethal trifecta applies to the agent on your laptop. Three questions to ask before you add the next MCP server.
- 2 notes, newest firstAll notes
Follow along
No newsletter. No pop-ups. Just RSS.
Every post and note, in full, in the feed reader you already use. No email address and no tracking pixel.
- Postschadnorman.fyi
/rss.xml - Noteschadnorman.fyi
/notes/rss.xml