How Her's AI Dream Stalled in Consumer Apathy

Major AI labs and venture capitalists have poured billions into a future dominated by smart, autonomous AI agents, yet outside Silicon Valley, nobody's using them. Wired reports that while OpenAI's Codex and ChatGPT Work agents have about 10 million weekly users, that figure is "basically a rounding error" compared to the general-purpose chatbots like ChatGPT and Gemini, which see around a billion monthly users. Josh Miller, CEO of The Browser Company, put it bluntly in a viral post this week: "Theoretically, the tech is ready... but alas the general public dgaf (doesn't give a fuck)."

The industry's core assumption is that consumers want powerful, autonomous AI that can book flights and write code, a vision often directly inspired by the movie Her. The data shows they don't. The gulf between a demo's promise and daily, reliable utility is a chasm most agent projects cannot cross. The problem isn't the models, it's the mindset. AI agents are not a product; they are a technology in search of a problem people actually have.

The Obvious AI Agent Is a Ghost Town

The disconnect is crystallized in a single developer's anecdote. According to additional context from roborhythms.com, a builder showed their sophisticated agent system to a friend who works at a bank. The friend watched the demo, nodded, and said: "So it's like a macro that sometimes works?"

"That sentence describes the AI agent market in 2026 better than anything I've read in a press release," the article notes.

That's the state of play. There is no killer consumer agent because builders are obsessed with shipping "the most impressive things their models can do," like navigating websites autonomously or chaining API calls. Miller argues these capabilities feel more like demos than daily products. He describes a viral groupthink within AI labs: "I think every single lab but one mentioned the movie Her as the way they articulated their vision."

The product that works is the one you don't think of as an agent at all. Miller's own company, The Browser Company, found success with a personalized morning briefing in its browser: a greeting, a to-do list pulled from a calendar, a piece of art. Technically, an agent powers it, but the user doesn't need to know that. This highlights the core thesis: the barrier isn't technological capability, it's a model-centric mindset that prioritizes technological novelty over solving known, mundane human problems.

The Real Numbers Behind the 'Agent' Illusion

The user metrics tell a stark story. While business-focused agents from OpenAI and Anthropic see adoption in the low millions of weekly users, mainstream chatbots operate on a scale 100 times larger. This isn't a niche success; it's a rounding error.

A broader look at AI adoption underscores the issue. According to Business of Apps data cited by XDA-Developers, ChatGPT, the most popular AI bot, has only 250 million active users weekly. In a global market with billions of connected devices, total regular AI usage sits around 300 million people. For context, that's less than half of Instagram's user base.

Where is the money going? Venture capital continues to chase the grand, autonomous agent vision, fundamentally betting that adoption will follow capability. But the data suggests a different trajectory: adoption follows reliability. The grand vision—an army of agents that replace human workflows—is underperforming. As the roborhythms.com analysis puts it, "autonomous AI agents are at or near peak inflated expectations," the precise stage where marketing hype collides with the hard wall of user indifference.

From Turing Tests to To-Do Lists: A Pivot in AI Ambition

This marks a significant, humbling correction in AI's ambition. The field's ultimate benchmark has long been the Turing Test or the creation of Artificial General Intelligence (AGI). Today's practical goal is far simpler: solve a specific, boring human problem without breaking.

We've seen this movie before. Earlier virtual assistants like Siri or Google Now promised proactive, ambient computing. They largely failed because they were unreliable and tried to do too much. Today's constrained agent goals—drafting an email for review, parsing a document—represent a retreat from those grand promises, focusing on being useful within a narrow band.

The industry is learning that intelligence for its own sake is not a product. Miller's argument pushes this further: the frame of an "AI agent" itself is an invented, insider term. "No one wants AI agents, because AI agents aren't a thing," he says. The winning application won't be announced as an agent. It will be a button in your email client that reliably schedules a meeting, or a feature in your bank app that helps you save, as we saw in our recent analysis of personal finance AI tools. The pivot is from building a brain to building a better tool.

The Stakeholder Split: Builders, Investors, and the Rest of Us

This adoption chasm exists because the three primary stakeholder groups are evaluating success by completely different criteria.

Builder Perspective: The technical fascination is intrinsic. Builders live in a world where an agent stalling at step three is an interesting debugging challenge, not a deal-breaker. They optimize for capability and autonomy.

Investor Perspective: There is immense pressure to hype the "next big thing," creating a reality distortion field. This hype cycle, like the one we've tracked in the robotaxi sector, can flood a sector with capital long before a real market exists, further insulating builders from user reality.

User Perspective: The demand is simple: reliable, trustworthy help. Finn McKenty summarized it on X: "People aren't using AI agents because: 1) They want to do LESS work, not more... 2) They're complex. Remember, most people have never even used 'SUM' in Excel... they're not going to set up an 'agent.'" Users evaluate on reliability, not potential. When something fails unexpectedly, they close the app and never return.

What a Truly Mainstream AI Agent Actually Looks Like

If we listen to users, the blueprint for a mainstream AI agent becomes clear, and it looks nothing like the sci-fi ideal.

Success is not autonomy; it's reliable, invisible assistance. The agent must disappear into an app people already trust, like their calendar, email, or banking software. Its presence is felt only through outcomes: a perfectly scheduled week, a summarized report, a filed expense.

Prioritize perfect execution of ten simple tasks over mediocre attempts at a thousand complex ones. The roborhythms.com analysis provides a sobering comparison of the promise vs. the production reality:

• What agents are supposed to do: Autonomously execute multi-step tasks.
• What happens in most production deployments: Stall or error at step 3-4, requiring human fix.

The pattern that actually wins is "extend the human," not "replace the human." The agent handles high-volume, low-stakes repetitive work. The human handles judgment, edge cases, and oversight. The winning architecture is one built for user trust and error recovery, not for maximum chain-of-thought length.

Rewiring the Industry's Instincts for the Next Phase

The path forward requires the industry to rewire its own instincts. The next wave of successful "agents" will likely be seen as disappointingly simple by today's builders—highly specialized tools that win by doing less, but doing it flawlessly.

This shift will benefit incumbents with distribution and deep user context—companies like Apple, Microsoft, and Intuit—over pure-play agent startups. They have the trust, the existing workflows, and the user relationships to integrate AI assistance seamlessly and reliably, much like infrastructure firms are building the necessary backbone for robotaxis.

The winning consumer agent won't be unveiled at a splashy keynote with a demo of it booking a complex multi-city vacation. It will be a feature you slowly realize you can't live without, like Miller's morning briefing. It will be so boringly useful that its underlying technology becomes irrelevant. As Miller concludes: "Who gives a shit if it looks like an AI agent? No one uses AI agents." The market is voting with its attention, and the result is a mandate for utility over wonder.

The Bottom Line

• Billions in investments risk being wasted on a technology that doesn't solve actual consumer problems.
• The AI industry's 'Her'-inspired vision of autonomous agents may be fundamentally misaligned with what everyday users need.
• Major labs are building impressive demos rather than reliable daily tools, creating a chasm between hype and adoption.