Human Centric Design in the Age Of AI

Disclosure: The first draft of the Article and initial sourcing were done with varying assistance of AI. The Article was then partially or entirely rewritten for publication standards and sources properly checked for accuracy

There is a version of this argument that gets written every few months, and it is almost always wrong. It goes: artificial intelligence is hollowing us out, the machines are coming for the last warm things we have, and the only honorable response is refusal. That essay is easy to write and easy to dismiss, because anyone who has actually used these systems knows the premise is false. AI is not a hollow trick. It is one of the most genuinely impressive things our species has built in a generation. The more interesting question, the one worth several thousand words, is not whether AI is brilliant. It is what happens to a civilization that gets very, very good at removing the need for one person to encounter another, and whether we are making that choice deliberately or simply drifting into it one convenient default at a time.

Start with the brilliance, because it deserves to be stated plainly rather than conceded grudgingly. In 2020, DeepMind's AlphaFold2 effectively solved a problem biologists had been grinding against since the 1970s: predicting a protein's three-dimensional structure from its amino acid sequence. Within two years the team had released predicted structures for over 200 million proteins very nearly every protein catalogued by science and made them freely available. The work was recognized with the 2024 Nobel Prize in Chemistry, and the database has since been used by millions of researchers across more than 190 countries.[^1][^2] That is not a productivity gimmick. That is a compression of decades of laboratory labor into an afternoon, handed to the entire world at no cost. Anyone arguing that AI is fundamentally a parlor trick has to explain that away, and they can't.

The gains are not confined to frontier science, either. In one of the first large field studies of generative AI at work, Brynjolfsson, Li, and Raymond tracked the rollout of an AI conversational assistant across roughly 5,000 customer support agents and found a 14% average increase in issues resolved per hour with a 34% improvement among novice and lower-skilled workers, and minimal effect on the most experienced ones.[^3] Read that carefully, because it is the most humane finding in the whole AI literature: the tool functioned as a leveler. It captured the tacit know-how of the best workers and handed it to the newest ones. Customer sentiment improved. Retention improved. This is technology doing what we always claimed we wanted technology to do.

So let us define our terms, and let us be fair to both sides of the definition. Human-centric design is the practice of building things by humans, for humans, that are meant to be used among humans. The lineage runs through Don Norman's foundational work on human-centered design in the 1980s, which framed design as an act of communication requiring deep understanding of the person on the other end and which Norman himself has since expanded into what he calls "humanity-centered design," widening the frame from the individual user to society and ecosystem.[^4][^5] Note what is embedded in all of it: a person is the endpoint. The artifact exists to serve someone, and very often to connect them to someone else.

Agentic design is something structurally different, and it is not evil, it is just aimed elsewhere. In agentic design, the primary consumer of the interface is a machine acting on your behalf. Its measure of success is delegation: how much of a task can be handed off, how little of your attention it costs, how few steps stand between intent and outcome. The ideal agentic experience is one where you state a goal and the world quietly rearranges itself. There is real dignity in this. It gives time back to people who have none. It gives capability to people who lack expertise. It removes drudgery that no one ever wanted in the first place. Anyone who has watched a caregiver lose an evening to insurance paperwork should want more of it, not less.

But here is the distinction that I think matters more than any other, and it is the hinge of this entire piece: human-centric design optimizes the quality of an encounter. Agentic design optimizes the elimination of the encounter. Those are not opposing moral positions; they are opposing objective functions. And when you optimize hard against a variable, you get less of it. The friction that agentic design correctly identifies as waste, the phone call, the counter, the awkward small talk with the person handing you the thing, is, in aggregate, also the substrate of social life. Some friction is grit in the gears. Some friction is the gears. A design ethos with no vocabulary for the difference will sand down both.

We have twenty years of data on what happens when you sand it all down, and it predates the current AI wave by decades, which is precisely why this is not an anti-AI argument. In his 2025 Atlantic cover story "The Anti-Social Century," Derek Thompson pulled the key number out of the Bureau of Labor Statistics' American Time Use Survey: from 2003 to 2023, in-person socializing fell by more than 20 percent, and among unmarried men and people under 25 the decline exceeded 35 percent.[^6] The trend began before the pandemic and continued after it. Thompson's argument is not that a single villainous technology did this. It is that each successive convenience, the car, the television, the smartphone, the delivery app, privatized one more slice of life that used to require leaving the house and standing near a stranger. AI did not start this. AI is arriving as the most capable tool yet built for continuing it.

And the stakes are not merely aesthetic. In 2023, the U.S. Surgeon General issued a formal advisory declaring loneliness and social isolation a public health crisis, situating them as independent risk factors for cardiovascular disease, dementia, depression, and premature mortality, and calling for the same order of national investment we have made against tobacco and obesity.[^7][^8] Whatever you think of any individual policy recommendation in that document, the underlying epidemiology is not in serious dispute: social connection is load-bearing for human health. Which means design decisions that systematically reduce the number of times a person is in the room with another person are, in the most literal sense, health decisions. We do not currently treat them that way.

What about AI specifically? The most careful evidence so far is nuanced enough that both camps should be uncomfortable with it. In a four-week randomized controlled trial run by the MIT Media Lab with OpenAI, roughly 1,000 participants across text and voice modes, light use of a chatbot was associated with reduced loneliness for some users, particularly with the emotionally expressive voice interface. But higher daily usage across all modes and conversation types correlated with greater loneliness, greater emotional dependence, more problematic use, and lower real-world socialization. A small subset of heavy "power users" accounted for a disproportionate share of emotionally loaded interactions.[^9][^10] The researchers are appropriately cautious about causality, lonely people may simply seek these tools out. But the dose-response shape should give any designer pause. A little substitution soothes. A lot of substitution appears to displace.

There is also an infrastructural version of this problem that is easier to measure and harder to argue with. Imperva's 2026 Bad Bot Report found that automated traffic accounted for more than 53% of all web traffic in 2025, up from 51% the year before, with human traffic down to 47% and still falling.[^11][^12] Set aside the security framing for a moment and sit with the raw fact: the internet is now, by volume, majority non-human. When most of your traffic is machines, the rational business response is to design for machines, clean APIs, structured data, agent-legible flows, and the human visitor becomes a rounding error you tolerate rather than a person you serve. That is how a design ethos dies. Not by anyone announcing its death, but by the slow reweighting of who the software is actually for.

None of this argues for refusing the tools. It argues for keeping a distinction alive that the market has no particular incentive to keep alive on its own. The practical version is something like: delegate the drudgery, not the relationship. AI that drafts the insurance appeal so a nurse can spend the hour with a patient is human-centric. AI that becomes the patient's only conversation is not, even though both use the same model, and the second one probably has better engagement metrics. The test is not whether AI is present in the system. It is whether the system's success is measured in encounters preserved or encounters avoided. Build the agent to clear the path to the other person, not to stand in for them. Instrument for it, too: if your dashboards only measure time saved, you will optimize away the things that were never supposed to be fast. Count the humans, not just the tasks.

The honest closing is that this is not a fight anyone wins by picking a side. AI is going to get better, agents are going to get more capable, and most of that is going to be good. But "human-centric design" only survives as an ethos if some people keep insisting it is a distinct thing worth naming, that there remain categories of product where the entire point is that another person is on the other end, and where a perfectly efficient machine substitute would count as a failure rather than a success. A world where every remaining transaction is mediated, brokered, and absorbed by an agent is not a dystopia. It is something quieter and harder to notice: a world where you never quite have to talk to anyone again, and gradually stop noticing that you don't. Whether we build that world is not a question about AI's capabilities. It is a question about what we choose to keep asking humans to do for each other, and whether we do it deliberately or by default.

Sources

[^1]: Nobel Prize in Chemistry 2024 press release, NobelPrize.org https://www.nobelprize.org/prizes/chemistry/2024/press-release/

[^2]: "AlphaFold: Five Years of Impact," Google DeepMind https://deepmind.google/blog/alphafold-five-years-of-impact/ (see also EMBL: https://www.embl.org/news/science-technology/alphafold-wins-nobel-prize-chemistry-2024/)

[^3]: Brynjolfsson, E., Li, D., & Raymond, L. (2023). "Generative AI at Work." NBER Working Paper No. 31161 https://www.nber.org/papers/w31161

[^4]: Norman, D. (2023). Design for a Better World: Meaningful, Sustainable, Humanity Centered. MIT Press https://mitpress.mit.edu/9780262548304/design-for-a-better-world/

[^5]: Norman, D. "Humanity-Centered versus Human-Centered Design" (excerpt, Ch. 22), jnd.org https://jnd.org/humanity-centered-versus-human-centered-design/

[^6]: Thompson, D. "The Anti-Social Century," The Atlantic, January 8th 2025. Underlying data: Bureau of Labor Statistics, American Time Use Survey https://www.theatlantic.com/magazine/archive/2025/02/american-loneliness-personality-politics/681091/

[^7]: Office of the Surgeon General (2023). Our Epidemic of Loneliness and Isolation. U.S. Dept. of Health and Human Services https://www.ncbi.nlm.nih.gov/books/NBK595227/

[^8]: AHRQ Academy summary of the Surgeon General advisory https://integrationacademy.ahrq.gov/news-and-events/news/surgeon-general-advisory-epidemic-loneliness-and-isolation

[^9]: Fang, C. M., Liu, A. R., Danry, V., Lee, E., Chan, S. W. T., Pataranutaporn, P., Maes, P., Phang, J., Lampe, M., Ahmad, L., & Agarwal, S. (2025). "How AI and Human Behaviors Shape Psychosocial Effects of Chatbot Use: A Longitudinal Randomized Controlled Study." arXiv:2503.17473 https://arxiv.org/html/2503.17473v1

[^10]: MIT Media Lab / OpenAI, "Early methods for studying affective use and emotional wellbeing in ChatGPT" https://www.media.mit.edu/posts/openai-mit-research-collaboration-affective-use-and-emotional-wellbeing-in-ChatGPT/

[^11]: Imperva, 2026 Bad Bot Report: Bad Bots in the Agentic Age https://www.imperva.com/blog/bad-bot-report-2026-bots-agentic-age/

[^12]: Imperva, 2025 Bad Bot Report (prior-year 51% figure for comparison) https://www.imperva.com/blog/2025-imperva-bad-bot-report-how-ai-is-supercharging-the-bot-threat/