The Brian blog

Field notes on safe delegation

Practical writing on giving AI agents your company’s judgment: procedures they follow, lines they don’t cross, and what to do about everything in between.

Playbook7 min read

SOPs for AI agents: turning company processes into rules agents actually follow

AI agents can’t follow SOPs written for humans. Here’s how to turn a process into a trigger, hard rules, and stop conditions an agent can execute.

Read the article
Playbook6 min read

AI agent guardrails: how to stop an agent before it does damage

Hard rules, stop conditions, and a tool risk registry: a practical guide to guardrails that stop an AI agent before the action, not after the incident.

Read the article
Opinion6 min read

Why system prompts are the wrong place for your company’s rules

System prompts duplicate, drift, and can’t say no. Why company rules belong in a versioned skill store agents query at task time, and how to migrate.

Read the article
Comparison5 min read

AI agent memory vs. governed skills: recall is not permission

Memory layers like Mem0 and Zep recall facts. Governed skills enforce rules. What each does well, where recall stops, and when you need both.

Read the article
Pattern6 min read

Can you trust an AI agent with refunds? The escalation pattern that makes it safe

A $40 refund gets processed. A $350 one gets refused, even when "the founder" insists. The bounded-delegation pattern that makes agents safe to trust.

Read the article
Explainer5 min read

What is MCP? The Model Context Protocol, explained for business teams

MCP is the port that connects AI agents to your tools. A plain-English guide to the Model Context Protocol, and the judgment gap it leaves open.

Read the article