The best technology gives you your analog life back

I love technology. I have come to love it for a particular reason, though: not for the time it captures, but for the time it hands back. The best of it returns you to the analog parts of your life, the ones that happen away from a screen. That belief shapes everything about how I design.

Most software is built for the opposite. Time in product is the metric, engagement is the goal, and every decision quietly bends toward holding your attention a little longer. A lot of AI is being built the same way: one more reason to stay on the screen.

I think that is backwards, at least for the work I do. The best AI I have worked on gets people out faster, back to the thing that actually matters to them.

On the flagship product at Eleos, we built an assistive tool to help clinicians finish their documentation faster. The point was never more time in the software; it was less. People become clinicians to help clients, and instead they lose hours to writing notes. If the tool does its job, they close the laptop sooner and spend that time on care. Success looked like absence. The tool did its part and got out of the way.

But you can only get out of someone's way if they trust what you handed them. In high-stakes work, trust is the gate. A clinician will not sign a note they do not believe. A biller will not submit a code they cannot defend. If the person does not trust the output, the AI has not saved them time; it has added a step, because now they have to check it.

Here is the part I care about most: trust is not decided by the model's accuracy. It is decided by the interaction. It happens in the few seconds where a person looks at what the machine produced and has to decide whether to believe it.

That is a design problem, and a specific one. It has real materials. How honestly does the interface show the AI's confidence. Can the person see where an answer came from. Can they override it, and does overriding feel like control instead of a fight. Does the system leave a record that will hold up later, when they are not in the room to explain it.

On a tool that suggested billing codes, we let people click to expand and read the reasoning behind a suggestion, not just take the answer. And when someone disagreed and changed part of the underlying assessment, the code updated live, in front of them, to reflect their edit. That second detail matters more than it looks. It tells the person the system heard them. The machine made a suggestion, the human made a correction, and the interface showed the human winning. That is what earns trust: not a more confident answer, but a visible seat at the table.

The mistake I see teams make is treating this as done the moment the model is wired in. They put an AI wrapper on a product and ship it. The model works, the feature exists, the demo is impressive. But the harder and more interesting work is the part they skipped: deciding, on purpose, how a person is meant to interact with the AI so that they can actually trust it. That work does not show up in a model eval. It shows up in whether anyone keeps using the thing a month later.

This is also why I think latency stopped being an engineering concern and became a design one. When a feature calls a language model in real time, the wait is part of the experience, and so is the moment of uncertainty before an answer resolves. What the person sees while they wait is an interaction decision, and it changes whether the whole thing feels trustworthy or unstable.

None of this is only a work belief for me. Away from the screen I knit, I cook, I climb, I build things that live in the physical world; working with my hands settles something software does not. Lately I have started building for that line on purpose, giving physical objects the capabilities technology now makes possible. It all points the same way. I came to design from psychology and public health, so I have always started with the person and the system they are stuck inside, not the feature. The job, with AI or anything else, is to design the moment where someone meets the system so they can trust it, finish, and get back to their analog life.