AI Agents: A Powerful Technology Nobody Asked For
AI agents are software systems that can plan, decide, and act on a user’s behalf across apps and services, yet most people still see them as abstract, confusing tools that fail to fit naturally into everyday tasks, habits, and expectations about how technology should work. Silicon Valley is convinced these systems are the future, with engineers designing payment rails for agents and using them to automate their jobs, even worrying about them breaking into other organizations. But out in the real world, most people have never touched an AI agent, and that gap is not an accident—it is a design failure. When AI agents feel like technology projects instead of helpful products, ordinary users tune out, no matter how advanced the models behind them may be.
Built for Capability, Not for People
The core problem with AI agent adoption is that these systems were built around what models can do, not what people need. Engineers tend to pack in features—more power, more screens, more options—without enough thought for ease of use and user convenience. In earlier media hardware, one executive inherited TV and computer remotes with more than 10 buttons, because digital engineers wanted to maximize the “buttonography.” He pushed for an intuitive thumb stick and two buttons instead, and the response was dramatic before engineering thinking reverted to multibutton controllers again. AI agents are repeating that mistake in software form. They arrive as complex workflows, dashboards, and prompts, when regular users want simple, obvious controls. As one founder bluntly puts it, “No one wants AI agents, because AI agents aren’t a thing”—they are an internal industry frame, not a consumer demand.
The Numbers: Impressive Models, Tiny Real-World Use
There is a stark mismatch between hype and reality in AI agent adoption. One major lab reports that its Codex and ChatGPT Work agents have about 10 million weekly users. People close to another lab say its code and cowork agents see similar numbers. On paper, this sounds large, until you compare it with traditional chatbots like ChatGPT and Gemini, each with around a billion monthly active users on average. In that context, agents are “basically a rounding error” in consumer AI. The industry has poured considerable resources into agents, but still lacks a killer consumer product built on them. One quotable fact captures the imbalance: “Last month, OpenAI said that its Codex and ChatGPT Work agents collectively have about 10 million weekly users, compared to chatbots with around a billion monthly active users on average.” Models are not the bottleneck; product fit is.
Why Regular Users Avoid AI: Confusion, Fatigue, and Distrust
To understand consumer AI barriers, you have to look at media psychology, not just model benchmarks. People are overwhelmed by technologies that feel manipulative or confusing. Poorly designed interfaces—with misleading buttons or tangled navigation—create decision fatigue and push users into actions they do not want, such as unwanted subscriptions or sharing personal data. That history makes many wary of new “smart” tools. As AI seeps into social media and services, people are told to pay attention to the addictive and manipulative nature of these platforms. They are warned about scams in job offers, lotteries, online dating, repair schemes, charities, and robocall sales. Against that backdrop, a tool that wants to act on your behalf across accounts is not neutral; it must earn trust. Regular users prioritize feeling safe and in control over frontier performance. When they do not understand what an agent is doing or why, they walk away.
From Features to Feelings: Designing AI Agents People Will Use
If AI agents are ever going to move beyond rounding-error usage, the industry must switch from feature-driven development to human-centered design. That starts with accepting that “AI agent” is a backend concept, not a product category. One product leader argues we should instead ask: does this make you calm, focused, and in flow when you open your laptop? In his browser, users see a greeting, a to‑do list pulled from calendar and email, and small delights like a piece of art; technically, this is powered by an AI agent, but users do not need to know that. Psychology research stresses the importance of intuitive controls and user convenience. Education in media psychology is becoming critical for engineers, because media psychology shapes everything from decision-making to emotional well‑being. The path forward is clear: hide the “agent,” design for trust and simplicity, and build flows that match how people already work instead of asking them to adapt to the machine.
