Writing
Engineering notes, with the evidence left in
Practical writing about agentic AI at work, evaluating AI systems, and systems analysis.
05 FIELD NOTES
2026
ARCHIVE
PHP2Shell and the smarter-attacker test: which defences still hold when AI finds the bugs
AI-assisted research found two critical PHP flaws. Which defences still work when the attacker is smarter than your whole team?
Before you connect another MCP server: an audit checklist for workplace teams
What MCP is, how a tool call works after the July 2026 revision, and seven questions to ask before an MCP server touches your work systems.
Badge or evidence: what a deployed-agent engineer should be able to show
Anthropic plans to certify 10,000 Frontier Deployed Engineers. What Canadian FDE postings already ask for, and the evidence to build now.
Securing AI agents in your infrastructure: a sandbox is only the first layer
Nvidia's OpenShell limits what an agent can reach. OpenAPPA limits where its data can go. Why agents inside your network need both, and more.
OpenAI's dots need acceptance criteria, not just Custom Rules
Custom Rules say what OpenAI's dots may do, not what a good result looks like. A systems analyst's case for earning agent autonomy through evidence.