Writing
What I learn from building AI agents and products, and what it means for how AI should be governed.
Follow by RSSAgents plan, call tools and act. Governance written for advisory software cannot stop an action — so authorization, approval and audit have to live inside the runtime.
Why frameworks written for advisory AI stop working once systems start taking actions — and four governance moves that survive contact with a real deployment.
A practical six-step guide — scoping the task, choosing tools, writing the loop, memory, human approval, and where governance actually fits.