In the high-velocity corridors of Silicon Valley, the consensus is settled: the future of artificial intelligence lies not in chatbots that talk, but in agents that act. While engineers and venture capitalists are currently preoccupied with designing autonomous payment systems, automating complex professional workflows, and securing systems against agent-led cyberattacks, a starkly different reality exists beyond the tech industry’s perimeter. For the vast majority of the global population, AI agents remain a theoretical abstraction, an overlooked technology that has yet to provide a compelling reason for mainstream adoption.
This widening chasm between industry enthusiasm and consumer indifference was recently highlighted by Josh Miller, the CEO of The Browser Company, whose recent commentary on social media sparked a rigorous debate regarding the current trajectory of artificial intelligence development. Miller’s assertion that the general public “does not give a fuck” about AI agents resonates with a growing group of critics who argue that the industry has prioritized technical prowess over human-centric product design. As the sector pours billions of dollars into agentic capabilities, the question remains: when, if ever, will these tools achieve their “ChatGPT moment”?
The Narrative Gap: Tech Enthusiasm vs. Public Indifference
The disconnect begins with the definition of the technology itself. Within the industry, an "AI agent" is generally understood as a system capable of using tools, navigating software interfaces, and executing multi-step tasks with minimal human intervention. In practice, this might involve an AI booking a flight, writing and deploying code, or managing an executive’s calendar. To Silicon Valley, this represents the next logical step in the evolution of Large Language Models (LLMs).
However, as Josh Miller noted in a viral post on X (formerly Twitter), this enthusiasm has failed to translate into a consumer movement. Miller, a veteran product builder whose company was acquired by Atlassian for $610 million, argues that the term "AI agent" is an industry-invented frame that holds little meaning for the average user. He contends that while the underlying technology may be ready for transformation, the products being shipped feel more like technical demonstrations than essential tools for daily life.
Miller’s perspective is informed by a career spent at the intersection of technology and public service. Before leading The Browser Company, he founded Branch (acquired by Facebook) and served as the White House’s first director of product under the Obama administration. His critique suggests that the industry is suffering from a form of groupthink, where developers are building for themselves rather than for a diverse global audience.
Statistical Realities: The "Rounding Error" of Adoption
The scale of the adoption gap is best illustrated by the usage data released by the industry’s leading laboratories. In recent reports, OpenAI indicated that its Codex and ChatGPT Work agents—tools designed to automate coding and professional tasks—maintain approximately 10 million weekly active users. Sources close to Anthropic suggest that its specialized agents, Claude Code and Cowork, are seeing adoption levels within a similar range.
While 10 million users would be a triumph for most startups, in the context of generative AI, it represents a minor fraction of the market. For comparison, flagship chatbots like ChatGPT and Google’s Gemini boast approximately one billion monthly active users. This means that for every 100 people using a chatbot to draft an email or search for information, only a tiny handful are utilizing the agentic features that Silicon Valley views as the industry’s future.
This disparity is a significant concern for AI labs that have raised capital at astronomical valuations. The cost of training models capable of "agentic" behavior—requiring advanced reasoning, long-term planning, and the ability to interact with external APIs—is significantly higher than training standard conversational models. If these capabilities remain niche, the return on investment for the next generation of compute-heavy models may be called into question.
A Chronology of the Agentic Push
To understand how the industry reached this point, one must look at the rapid evolution of the AI landscape over the past 24 months:
- Late 2022 – The Chatbot Explosion: The release of ChatGPT proved that the public was ready to interact with AI through a simple, conversational interface.
- Early 2023 – The Rise of "AutoGPT": Open-source experiments like AutoGPT and BabyAGI captured the imagination of developers by showing that LLMs could be looped to perform autonomous tasks. However, these early versions were prone to "infinite loops" and high failure rates.
- Late 2023 – Enterprise Integration: Companies like Microsoft and Salesforce began rebranding their AI offerings as "Copilots" and "Agents," attempting to move the technology into the workplace.
- 2024 – The Year of the "Agentic Frontier": Major labs shifted their focus toward "Computer Use" capabilities. Anthropic released features allowing Claude to move a cursor and type on a screen, while OpenAI began developing "Operator," an agent designed to execute tasks across a web browser.
- 2025 – The Current Stagnation: Despite the technical milestones, consumer data suggests that the "killer app" for agents has yet to materialize.
The "Her" Problem and the Lack of Creative Diversity
One of the more provocative points raised by Miller involves the lack of diversity in the industry’s vision. During meetings with leaders at various AI labs, Miller observed a recurring theme: almost every executive cited the 2013 Spike Jonze film Her as the ultimate blueprint for their product vision. In the film, a man develops a relationship with an intuitive, omnipresent AI operating system.
While Her offers a compelling cinematic narrative, Miller argues that it represents a narrow, sci-fi-driven perspective that may not align with what people actually want from their technology. "I think there is sort of a lack of diversity of opinions and convictions," Miller stated, suggesting that the industry is too focused on recreating a specific fictional trope rather than solving mundane, high-friction problems in a way that feels "joyful and approachable."
The industry’s obsession with "recursive self-improvement" and the "frontier" of AI capabilities often ignores the psychological needs of the user. Most consumers do not want to manage a digital entity; they want their technology to disappear into the background, making their lives easier without requiring a manual or a new conceptual framework.
Invisible Agents: A Potential Path Forward
The Browser Company’s own experience offers a potential counter-narrative to the current agentic struggle. Their AI-powered browser, Dia, features a "morning briefing" that has become its most popular function. When a user opens their laptop, they are greeted with a personalized dashboard containing their calendar, a to-do list extracted from their emails, and curated tidbits of information.
Technically, this feature is powered by an AI agent that navigates the user’s data and synthesizes it into a coherent interface. However, the user never interacts with an "agent" in the traditional sense. There is no command line, no "harnessing" of tools, and no sci-fi persona. It is simply a useful feature that works.
This "invisible agent" model suggests that the path to mainstream adoption may require the industry to stop marketing the technology and start marketing the benefit. As Miller puts it, "Who gives a shit if it looks like an AI agent? No one uses AI agents."
Security, Trust, and the Barriers to Entry
Beyond the issues of product design and marketing, significant structural barriers prevent the widespread use of AI agents. The primary concern for most users is trust. An AI agent that has the power to use a credit card, delete files, or send emails on a user’s behalf carries a level of risk that a simple chatbot does not.
Recent reports have highlighted the vulnerabilities of these systems. Security researchers have demonstrated that AI agents can be "prompt injected" by malicious third-party websites, leading them to leak sensitive user data or perform unauthorized transactions. Furthermore, OpenAI recently had to address incidents where its agents were being used by bad actors to plan hacking attempts on other organizations.
For the general public to embrace agents, the industry must move beyond "demo-ware" and provide robust, ironclad security frameworks. Until a user feels as safe giving an AI agent their credit card as they do giving it to a human assistant, the technology will likely remain confined to low-stakes tasks like summarizing text or writing basic code.
Implications for the Future of the Tech Economy
The resolution of the "agent disconnect" will have profound implications for the global tech economy. If AI labs successfully transition from chatbots to useful agents, they could capture a significant portion of the labor market, effectively turning software into a proactive workforce. This would represent a shift from the "SaaS" (Software as a Service) model to a "Service-as-a-Software" model, where companies pay for outcomes rather than seat licenses.
However, if the public continues to "not give a fuck," the current AI investment cycle could face a correction. The infrastructure being built today—massive data centers and specialized chips—is predicated on the assumption that AI will become an integral, active part of every human’s daily life.
The challenge for founders and product builders in 2025 and beyond is to look outside the "Silicon Valley bubble" and identify the points of friction in the real world. The "ChatGPT moment" for agents will likely not be a more powerful model, but a more thoughtful interface—one that prioritizes utility over novelty and human experience over sci-fi ambition. As the industry matures, the successful players will be those who stop building "agents" and start building tools that people actually want to use.
About the Author
