How agents decide, and when they stop

What a real decision needs, who enforces each rule, and why an agent that can go anywhere has no reason to stop: six articles on deciding, for people and agents alike.

These six were written over two days and none of them sets out to be a series. Read together they make one argument: deciding well needs more than a yes or a no, and stopping well needs a perspective that has to be designed in, for agents even more than for people.

Agency is not a yes is the frame: seven things a decider needs, from options beyond yes to authority over the system that produced the request, and a scale on which the weakest of them caps the decision. Where is the why? is the same problem met in a permission prompt with no reason on it.

Knowing when to stop starts from people: security champions who had automated away whole classes of bug, then found themselves debating whether GUIDs were random enough. Agents have it worse, because their range is the point. Who are you protecting against? gives one way to stop: name the attacker before deciding how secure to be. Every mistake added a rule shows what happens without one: each mistake adds a rule, the rules become the complexity, and the way back is shipping small components instead.

Hope or enforcement is where the argument is tested: one customer service agent, three designs, and a count of which rules a model is only hoped to keep. The desk's thread note pulls out the line each of them shares.

The articles

2026-10-08Agents & policy7 threadsAgency is not a yes: a scale for human and agent decisions, from rubber stamp to the reviewer who fixes the sourceA yes or no is not agency: seven dimensions, a seven-level scale for people and agents, and the review as a QA step that fixes the source.

2026-10-08Agents & policy7 threadsKnowing when to stop: what experience gives people, and what we have to design into agentsKnowing when to stop is the hard part for people and agents: what experience gives people, and the constraints that give agents the same perspective.

2026-10-08Agents & policy8 threadsEvery mistake added a rule: complexity, agents, and the way back to shippingWhen every agent mistake adds a rule, complexity wins: map the process, move each piece right as a small shipped component, and keep the rigour for the work.

2026-10-08Agents & policy4 threadsHope or enforcement: one customer service agent, three designs, and who keeps each promiseOne customer service agent built three ways, each with an Agent Behaviour Policy: how much of each policy is hope, and what one run can reach.

2026-10-08Agents & policy3 threadsWho are you protecting against? Draw the security line where the attacker is, not above itBefore deciding how secure to be, name the attacker: a six-tier ladder, an air-gapped Mac mini read against it, and eight startups drawing their line.

2026-10-07Agents & policy12 threadsWhere is the why? A permission prompt asked me to decide, and kept the reasonA prompt asked for a decision and kept the reason. Read through the policy, the law on uninformed consent, and the fixes that worked: put the why in the prompt.

← All collections · The front page