Trust is a feature

I use ChatGPT and other AI products every day. They help me think through a problem, explore an idea or get something done faster.

But sometimes they also remind me how much my expectations of software have changed.

Software used to be something I could mostly trust to behave predictably. I gave it an input, followed a process and expected a consistent result.

AI changes that relationship. The same request can produce a different answer. A result can sound convincing and still be wrong. And sometimes it takes more effort to verify the output than it would have taken to do the task myself.

That does not make AI bad. From my point of view it makes one feature more important, a feature we rarely put on the list: Trust.

When we talk about features we usually name what users can see. Dashboards, integrations, workflows, automation. But whether people keep using a product often depends on simpler questions:

Does it behave consistently?

Can I understand what happened?

Can I recover when something goes wrong?

Can I rely on it when the stakes are high?

AI makes these questions more visible, because its outputs vary and its mistakes can sound convincing. But they apply to all software. A beautiful interface does not compensate for unreliable behaviour. A clever automation is not useful if users cannot tell what it did. A fast workflow is not valuable if it creates uncertainty at every step.

Trust comes from many small things: clear feedback, predictable behaviour where possible, sensible defaults, verification, good error handling and honest communication about limitations. It is not a finishing touch added after the product works. It is part of what makes the product work.

A great demo is not enough. A clever answer is not enough.

As we add more AI to software, I think we should spend at least as much time asking "Can people trust this?" as we do asking "Can we build this?"