If you’re building agentic systems for high-stakes consumer workflows, here’s a simple finding that has enormous payoffs (based on speaking to 100+ users):
- The more consequential the action, the more trust the product has to earn.
- That means a lot of the real work is trust building on the ground, then reflecting that trust in the product itself.
- Sometimes emphasising autonomy can make the product feel less trustworthy, not more advanced.
- But it doesn’t mean users want a human involved.
- They want confidence and confirmation that the system works.
There's an old UX principle for this called Nielsen's "visibility of system status.” The system should always keep users informed about what's going on, through appropriate feedback, within a reasonable time. It was written for buttons and loading spinners. But it applies just as much, maybe more, to agents, because the whole premise of an agent is that it's doing things for you, often out of view. The less visible the work, the more that principle matters
This means clear checkpoints matter:
- Being able to review something before it is submitted matters
- Knowing what the system is doing in the background matters
- Knowing upfront what happens when something goes wrong matters
- Simply feeling like there is a real, accountable company behind the product matters.
Imagine placing an Uber Eats order and, after checkout, the interface just said: Processing.
No confirmation that the restaurant accepted it. No estimated delivery time. No map. No indication that the driver picked it up. No way to see where things went wrong. The actual service could be working perfectly, but people would still feel anxious, and support requests would probably skyrocket.
Many high-stakes agentic products work like this today, but much of it can be solved in the interface.
- A status update.
- A checkpoint.
- A review screen.
- A clear explanation of what happens next.
- A visible escalation path.
Basically small UX decisions can dramatically change how much autonomy a user is willing to give a system. And platform trust compounds as users become comfortable delegating more autonomy.