AI Agents: Silicon Valley’s Obsession Struggles for Real-World Adoption

In the insular world of Silicon Valley, a profound conviction has taken root: AI agents represent the undeniable future of technology. Within the bustling heart of the tech industry, engineers and entrepreneurs are dedicating vast resources to developing sophisticated payment systems designed specifically for these autonomous entities. They are leveraging AI agents to automate complex job functions, streamlining operations across various sectors, and, perhaps most notably, are actively engaged in efforts to prevent these powerful agents from maliciously hacking into other organizations. Yet, despite this fervent belief and intensive development within the tech sphere, the vast majority of people in the "real world" remain largely untouched by AI agents, having never directly interacted with or even perceived the utility of such tools. This stark disparity suggests a critical oversight: tech companies may not yet have presented a compelling reason for the general public to embrace them.

This growing disconnect was sharply articulated by Josh Miller, the CEO of The Browser Company, in a viral post on the social media platform X (formerly Twitter) this week. "Hot take," Miller began, "isn’t it kinda crazy that nobody is really using AI Agents?" He continued, "Theoretically, the tech is ready for AI agents to totally transform how we work and live our lives… but alas the general public dgaf" – an acronym succinctly conveying "doesn’t give a fuck." Miller’s candid "agentic epiphany," as it was dubbed, struck a chord within Silicon Valley, where many are broadly bewildered by the public’s apparent indifference to what has become the industry’s latest obsession. Speaking in an interview, Miller explained that he impulsively shared his thoughts while half-asleep during a family vacation in Europe, yet he firmly stands by his statement. He contends that the industry’s path forward must pivot towards constructing agent-based products that genuinely resonate with consumer desires and needs.

"I just have not heard a single person outside of the tech community talk about an agent that they use," Miller emphasized, highlighting the echo chamber effect prevalent in the industry. "As excited and optimistic as we as an industry may be about the frontier and recursive self-improvement, it’s worth pausing and just saying, ‘Huh, what?’" This sentiment underscores a critical moment of introspection for an industry accustomed to rapid, widespread adoption of its innovations. The current landscape for AI agents reveals a significant gap between their theoretical potential and their practical integration into daily life.

To contextualize the scale of this adoption challenge, recent data offers a sobering perspective. Last month, OpenAI reported that its combined suite of Codex and ChatGPT Work agents had reached approximately 10 million weekly users. Sources close to Anthropic, a prominent AI research company, indicate that their Claude Code and Cowork agents are experiencing comparable levels of engagement. While these figures might seem substantial in isolation, they shrink dramatically when compared to the mainstream success of conversational AI platforms. Chatbots like OpenAI’s ChatGPT and Google’s Gemini boast an average of around one billion monthly active users each. In this light, the user base for AI agents amounts to what can only be described as a statistical "rounding error," barely registering on the broader generative AI landscape.

This disparity presents a considerable problem for the major AI research laboratories that have invested billions of dollars into developing and training increasingly powerful AI models. Currently, most consumers engage with generative AI for relatively straightforward tasks, such as querying information or engaging in simple conversational exchanges. However, the advanced models developed by these labs possess capabilities far exceeding these basic functions. AI agents, by definition, are designed to perform more complex, multi-step tasks autonomously, often interacting with various systems and tools to achieve a user-defined goal. They represent a crucial avenue for these companies to fully capitalize on their immense research and development investments – but only if they can achieve significant public adoption. The current lack of a "killer consumer product" for agents is a frequently voiced concern among AI insiders, who are keenly awaiting an equivalent of the "ChatGPT moment" that democratized access to generative AI. The question remains: when, if ever, will such a breakthrough occur for AI agents?

Miller’s arguments carry considerable weight, not least because of his impressive track record in the tech industry. Despite his self-deprecating humor – he often points out he is "just" a sociology major – his insights are rooted in deep experience. His most recent venture, The Browser Company, developed Arc, an innovative web browser that has garnered a dedicated "cult following" and was successfully acquired by Atlassian last year for a reported $610 million. Prior to that, his first company, Branch, was acquired by Facebook in 2014. Miller also holds the distinction of having served as the White House’s inaugural director of product under President Barack Obama, bringing a unique blend of technical acumen, entrepreneurial success, and public sector experience to his current critique.

Miller’s central thesis extends beyond the mere observation that AI agents haven’t yet achieved mainstream traction. He argues that the fundamental framing of "AI agents" as a distinct product category is flawed. "No one wants AI agents, because AI agents aren’t a thing," Miller asserts. "It is an invented frame made up by our industry to collectively refer to something." His point is that users don’t inherently desire a technology; they desire solutions to their problems, improved experiences, or new capabilities that enhance their lives. The underlying technology, whether it involves a complex "harness" that calls up various tools or intricate recursive self-improvement algorithms, should be invisible to the end-user. The focus, he contends, should be on crafting compelling products that deliver tangible benefits.

He offers a concrete example from his own company. Miller highlights that The Browser Company’s most popular feature within its AI-powered Arc browser, named Dia, is a personalized morning briefing. When users open their laptops, Dia presents them with a custom homepage that includes a warm greeting, a curated daily to-do list intelligently populated from their calendar and email, and an assortment of "random tidbits" – such as a piece of art or an interesting fact – designed to spark joy and creativity. Technically, Miller explains, this seemingly simple yet highly engaging feature is made possible only through the sophisticated orchestration of an AI agent operating behind the scenes. Crucially, however, the user is entirely unaware of this underlying technological complexity. Their experience is one of calm, focus, and a seamless transition into their workday, precisely the product Miller advocates for.

While it is convenient for Miller to champion AI-powered web browsers, given that he sells one, his critique of the current agentic product landscape appears valid. Many of today’s agent-focused offerings are indeed not designed with the average consumer in mind. Tech companies, in their eagerness, often showcase the most impressive and technically challenging feats their models can achieve – such as autonomously navigating a website, generating complex code, or orchestrating multi-step workflows. These demonstrations, while technologically advanced, often feel more like proofs-of-concept or elaborate "demos" rather than fully realized products tailored to address specific customer desires or pain points. They highlight capability over utility.

Miller attributes a significant portion of this problem to what he describes as "groupthink" within the AI industry. He observes that many individuals actively involved in building these cutting-edge technologies are deeply immersed in and often singularly obsessed with them. This immersion frequently leads to a shared, almost uniform, sci-fi-inspired vision for how these products should manifest and interact with users. Miller recounts a striking anecdote from the previous summer, during which The Browser Company explored potential acquisition opportunities. He met with numerous leaders from top AI labs, and the commonality was startling. "It was wild to me," Miller stated. "I think every single lab but one mentioned the movie Her as the way they articulated their vision." He acknowledges respect for those who pursue their singular vision, but cautions that such a lack of "diversity of opinions and convictions" can narrow the scope of innovation and prevent the creation of products that appeal to a broader audience. The Her vision, while romanticized in fiction, often implies a deeply personal, almost sentient AI companion, which may not align with practical, everyday needs or consumer comfort levels.

The current prevailing trend within the AI agent space involves empowering users to construct personalized software solutions to automate specific aspects of their lives. This includes tasks such as generating professional pitch decks, organizing intricate datasets, or streamlining other bespoke productivity workflows. Indeed, tools like OpenAI’s ChatGPT Work and Anthropic’s Claude Cowork are proving increasingly useful for those who engage with them. However, Miller points out that the demographic interested in actively building and customizing their own productivity tools remains a relatively limited pool. For AI agents to truly achieve mainstream adoption and fulfill their transformative potential, the industry must develop more expansive and universally appealing ideas that transcend niche applications.

Miller hopes his urgent message will resonate beyond the confines of OpenAI and Anthropic, reaching a broader audience of founders, product managers, and developers across the tech ecosystem. His overarching goal is to encourage a fundamental shift in perspective, pushing people to think creatively and "outside the box" about the myriad forms that AI agent products could potentially take, and then to diligently bring those novel visions to fruition.

"Let’s not just accept this narrative of AI agents and actually question what are good products and tools that are joyful, useful, and approachable," Miller urged. "Who gives a shit if it looks like an AI agent? No one uses AI agents." His call to action is clear: the industry must prioritize user experience and genuine utility over technological prowess or adherence to a predetermined, industry-internal label. The true success of AI agents will lie not in their technical sophistication, but in their ability to seamlessly integrate into and genuinely enhance the daily lives of the general public, often without them even realizing the powerful AI agent working behind the scenes.

This report is an edition of Maxwell Zeff’s Model Behavior newsletter.