Technology / AI & Machine Learning2 min read
AI agents no longer just answer. They act. Before we let them book, buy or send on our behalf, we need to design for trust.
There is a moment every designer of AI agents will face. The agent has done the research, filled in the forms and is one click away from booking the trip, sending the email or moving the money.
Should it just go ahead?
AI agents have moved from demos into real products. They browse, compare, write and act on our behalf. That is powerful. It is also a completely different relationship between people and software. We are no longer only asking users to operate a tool. We are asking them to delegate.
And delegation requires trust.
The principles that help here are not new. Jakob Nielsen’s first usability heuristic, visibility of system status, says “The design should always keep users informed about what is going on, through appropriate feedback within a reasonable amount of time.”
For an agent that works for minutes instead of milliseconds, that heuristic becomes essential. Silence feels like something has gone wrong.
His fifth heuristic, error prevention, is just as relevant. “Either eliminate error-prone conditions, or check for them and present users with a confirmation option before they commit to the action.”
Here is how I translate those principles into agentic UX.
- Show progress in plain language. Not a spinner. A short description of what the agent is doing right now and what comes next.
- Make the plan visible before it runs. Let people review the steps and adjust them before anything happens.
- Add checkpoints before irreversible actions. Payments, deletions, sent messages and bookings deserve a clear pause and a clear summary.
- Keep an emergency exit. Users should be able to stop, pause or take over at any point.
- Leave a trail. A simple log of what the agent did builds confidence and makes mistakes easy to trace.
Autonomy is earned. It is never assumed.
Start with more checkpoints than you think you need. As people see the agent behave reliably, give them the option to remove some of them. Let trust grow at the pace of the user, not the pace of the product roadmap.
Brené Brown said it best in her talk The Anatomy of Trust. “Trust is built in very small moments.”
Every accurate progress update is one of those moments. Every well timed confirmation is another.
Design for trust first. Autonomy will follow when people are ready to give it.
