For Years, Big Tech Has Harvested Our Data. Now It Wants Us to Hand Over Our Lives

The leap from conversational assistants to agents isn’t an incremental improvement: it’s a whole new ball game. A chatbot responds; an agent acts. Enter your email and it schedules meetings, buys stuff, books concerts, reorganize documents, consults repositories, uses passwords, operates on services and, in short, does things on your behalf. That’s why today’s race isn’t simply about “making life more comfortable,” but about conquering the most intimate layer of digital life.

Meta is developing an agentic assistant for its more than three billion users capable of performing everyday tasks in return for their health and financial data. The source quoted in the article summed up the problem with a devastating sentence: there is a trust deficit “as wide as the Grand Canyon.” Meanwhile, Google is working on permanent assistants integrated into Gemini that can act on users’ behalf in all areas of their digital life.

As always, the official explanation is convenience. The agent buys for us, summarizes emails, organizes trips, finds documents or manages calendars. And yes, all of that can be extraordinarily useful. But let’s remember who the two main players here are: Meta and Google, two companies whose dominant position was built on the mass capture of personal data, behavioral inference, and the advertising monetization of our daily lives.

This is where OpenClaw, the catalyst for this race, comes in. It is an open agent platform that allows users to create their own agents to automate complex tasks. Its recent popularity is no accident: it showed that there is a real demand for systems that not only respond, but act. But it also revealed something much more interesting: this power can, at least in part, remain in the hands of the user.

Unlike the centralized solutions of Big Tech, OpenClaw is distributed as software that the user can run in their own environment, either on a computer or in a cloud space. That implies, with important nuances, that data, credentials, and operational context remain under your control, depending on how it’s configured and deployed. It’s not an absolute guarantee of privacy, of course, but it is a paradigm shift: the agent doesn’t necessarily have to be a corporate intermediary: it can be yours.

The problem is that OpenClaw is not for everyone. Installation, configuration and maintenance are still clearly out of reach of the average user. It requires technical knowledge, understanding permissions, integration with services and, above all, a certain culture of technological self-management. And that’s exactly where Meta and Google have seen the opportunity.

Meta obtained revenue of $200.7 billion from advertising, with more impressions and a higher average price per ad. Google remains, essentially, a gigantic advertising intermediation machine, now bolstered by AI and an ever-deepening integration of its services. When companies with that track record say they want to create “personal assistants,” the right question is not what tasks they will do for us, but what information they will capture from us while doing them. Because what these companies are trying to do is not just build agents, it’s eliminate friction. Turning something that today is complex, distributed and potentially controlled by the user into a simple, integrated, invisible service, and centralized. What Zuckerberg is proposing when he talks about “a polished version that just works” is not a usability improvement: it is a reappropriation of the model.

History repeats itself. The open web gave way to closed platforms because they were easier. Installable software was replaced by cloud services because they were more accessible. Self-managed mail gave way to Gmail because it worked better. And in each transition, the user gained ease and lost control.

Now the same thing is happening with agents. OpenClaw represents an embryonic version of what a decentralized ecosystem of personal agents could be, but its complexity makes it a niche product. Meta and Google intend to solve that problem… in exchange for turning the agent into another data extraction vector.

And the incentives are clear. Analysts quoted by CNBC put it bluntly: agents are a new revenue infrastructure, capable of generating value through commerce, advertising, and productivity. Moreover, they create engagement, increase dependency and reinforce lock-in. They don’t just know more about you, they do more for you. And the more they do, the harder it becomes to do without them.

An agent is the wet dream of surveillance capitalism. It not only observes what we publish or look for: it observes intentions before they materialize, it knows what doubt we have before deciding, what purchase we are considering before executing it, what problem we are trying to solve before formulating it completely. We’re no longer talking about profiling users from passive data: we’re talking about embedding an active system at the core of their behavior.

The risk is not only that the agent makes a mistake, although the examples already exist, such as agents that delete emails or execute unwanted actions. The risk is structural. To function, these systems need continuous access, broad permissions, and persistent memory. Each of these elements is a data capture surface. And each piece of data feeds models whose objective is not to serve the user, but to monetize their behavior.

Precedent alone should set alarm bells ringing. The same companies that now aspire to manage our actions have exploited our information with little transparency and multiple regulatory conflicts for years. To think that, in this new context, they will act as neutral guardians of our privacy is to ignore all the evidence accumulated over the years.

Agents are systems capable of reasoning, planning, using tools, remembering and acting to meet objectives, and that ability expands the risk surface beyond the traditional prompt injection. An agent connected to authenticated browsing such as Gmail, Google Drive or GitHub can exfiltrate data or do serious damage if it receives malicious instructions. The question is not whether there will be incidents: the question is how many, with what impact and under what liability regime. But as always, no one is going to read the terms of use, so it won’t matter.

The key word here is “agency.” Delegating to an agent is ceding the capacity for action. And doing so in favor of a platform whose business model depends on influencing our decisions introduces a fundamental conflict. When the agent decides what to buy, what to prioritize or what to suggest, it does so within an environment designed by a company with its own interests.

That is why the agency race is not a natural evolution of AI, but a battle for control of the most critical interface of all: the one that mediates between intention and action. Whoever controls the agent will control not only the data, but also the decisions.

Agentic agents are certainly convenient. So was the use of free social networks, free email or cloud storage. But it was never free. And now the price can be even higher, because we don’t just give up information, we give up operational capacity.

The alternative exists, even if it is imperfect today. It goes through agents that we can run locally, audit, control, and limit. For open standards, real interoperability and models that do not depend on surveillance. OpenClaw points in that direction, but its relative complexity still keeps it out of reach for most. And that is precisely why Big Tech sees a historic opportunity.

Meta and Google not simplifying technology out of altruism: they’re redesigning the checkpoint. They’re not competing to give us better tools, they are competing to make sure that when we delegate our actions to a machine, that machine belongs to them and they can control it. And at that point, surveillance will cease to be something that happens while we use technology, and will become something that acts for us at the service of a Big Tech. You have been warned.

—

This post was previously published on Enrique Dans’ blog.

***