Earned autonomy: how much should an agent do before it has your trust?

Every agent product hits the same question in its first month: how much should it do on its own?

Too little and it's a chatbot with extra steps. Too much and the first mistake costs you the user. The answer I keep coming back to is that autonomy shouldn't be a setting. It should be earned, the same way it is with a new colleague.

Nobody gives a new hire the company card on day one

Think about how trust works with a new person on your team. In week one they suggest. In month one they act on small things and tell you after. By month six they handle whole areas and only escalate the odd case. Nobody signs a form for each step. They earn it by being right, visibly, again and again.

Agents should work the same way. Not "autonomous: on/off", but a ladder the agent climbs with each user.

The ladder

1. Suggest. The agent says what it would do. The person does it.
2. Prepare. The agent does the work and stops one click short. The person approves.
3. Act and tell. The agent acts on small, reversible things and reports back straight away.
4. Act and summarise. The agent handles a whole category and sends a digest.
5. Act and escalate. The agent runs the area and only asks when something is unusual.

The product decision isn't which rung to build. It's what moves a user up a rung, and what moves them down.

What earns a rung

• A track record the user can see. "I've suggested this 12 times and you accepted 11" is a better reason to offer autonomy than any setting.
• Low cost of being wrong. Start autonomy where mistakes are cheap and reversible. Renaming a file before moving money.
• The user asking for it. "Just do it next time" is the strongest signal you'll get. Design for it.

What loses a rung

Trust doesn't fall gradually. As I wrote in The Yes Machine, it falls off a cliff. One confident wrong action and the user doesn't just drop a rung, they go back to doing it themselves and stop believing the agent's suggestions too.

So the ladder needs a way down that isn't the cliff: when the agent is unsure, it should step itself down a rung and say so. "I'd normally just do this, but this one looks different, can you check?" protects more trust than getting it right silently ever builds.

The agents that win won't be the most autonomous on day one. They'll be the ones people choose to give more to, a little at a time.