Commented on Agentivity · Nov 3, 2026
Great question, and honestly it’s the first thing I’d ask too before letting AI touch real work. Short answer: you stay in control, and you pick how strict you want to be. The light version is just telling the agent. Something like “ask me before spending more than $100”. If it has a way to reach you, it’ll stop and wait for your answer. It works well, but it’s still the model judging, so I wouldn’t rely on it alone for money. Then there are guardrails: you attach them to an agent and they check what it reads or writes. They can warn you, rewrite the text, or simply block it. And when it really matters, I’d go for a workflow. A plain condition (“over $100? go to approval”) always takes the same path, and an approval step waits for you every single time. No surprises. In the blog post example, that’s exactly the setup: the team writes the draft, then everything stops until you say yes. Nothing reaches WordPress or the editor before that. Happy to dig into your own case if you tell me what you’d want to automate!
