Why I Study HCI

Beyond optimization

There is a line from Robert Frost:

I shall be telling this with a sigh Somewhere ages and ages hence: Two roads diverged in a wood, and I— I took the one less traveled by, And that has made all the difference.

I keep coming back to it because of what I see in my own work. I do industrial design in e-commerce, and the honest read is that the field is stuck in optimization — squeezing conversion, shaving friction, iterating on the same funnel. That’s not nothing, but it isn’t growth. It’s the road most traveled, and I don’t think it’s where the next real work is.

I believe research on technology changes how society works, how business runs, and how people understand the world — and that belief is what’s pulling me off the optimization path. As AI becomes accessible to everyone, I think the discipline best positioned to make sense of that shift isn’t pure engineering. It’s HCI: the field that already sits at the intersection of social science, engineering, and natural science, built to ask what a technology does to the people using it, not just whether it works.

That’s the unconventional path, for me: choosing to study the interface instead of only optimizing it.

When agents do the work, alignment is the work

Here’s where that belief stopped being abstract. Uber Engineering recently described their internal state: more than 70% of pull requests now attributed to local or cloud agents, over 3,600 agent skills built across the software development lifecycle, and 30,000+ agent skill executions a day. A growing share of sessions aren’t started by a human at all — they’re managed agents handling code review, self-healing CI failures, completing end-to-end PRs with visual validation, triaging on-call alerts, debugging incoming bugs, with humans stepping in for review and escalation.

That last part is the part I care about. When agents handle execution, the human’s job shrinks to the review and escalation boundary — deciding what to trust, what to check, what to hand back. That boundary is a human-computer interaction problem, not a model-capability problem. From my own experience building with and around agents, I’ve felt how easy it is for that boundary to be badly designed — for “human in the loop” to mean a rubber stamp instead of real judgment. Getting that boundary right, the alignment between what agents do and what humans actually need to verify, is a design problem before it’s anything else.

AI is also lowering the barrier to doing research at all — more people can investigate hard problems than could five years ago. I think that raises, not lowers, the bar for what it takes to reach mid-senior roles in tech: the differentiator stops being “can you build it” and starts being “do you understand the people and the agents well enough to know when the output is right.”

I don’t know yet whether that’s a net good. Gresham’s Law says bad money drives out good — cheap, fast, agent-generated work could crowd out slower, more careful, well-aligned work if the review layer doesn’t hold. I’ve been in this field less than five years, so I’ll say plainly that I don’t have the answer to that. But it’s the open question I want to spend my research on.