The Desktop as a New Battleground
On March 16, 2026, Meta quietly activated one of its most significant investments in recent history. Manus, the artificial intelligence agent platform acquired in December 2025 for over $2 billion, launched its desktop application featuring a function no mass-market vendor had previously offered with such simplicity: direct control over the user’s local system.
The feature, named "My Computer," operates on macOS with Apple Silicon chips and on Windows. It is not an assistant suggesting what to do. It is an agent executing commands on your terminal, organizing thousands of files, processing batch invoices, and capable of building a functional translation application for meetings in under 20 minutes, without the user touching Xcode or any traditional development tools. The distinction from everything that existed before is not a matter of degree; it is a matter of nature.
To understand the weight of this, one must step back. Manus did not originate within Meta. It is a startup founded by Chinese entrepreneurs based in Singapore under the parent company Butterfly Effect Technology. It debuted in March 2025 and was labeled "revolutionary" by Forbes. A month later, it had already raised $75 million, led by Benchmark, with a valuation of $500 million. By December of the same year, Meta closed its acquisition at four times that figure. In less than nine months, the price multiplied by four. This appreciation rate cannot be attributed solely to user traction; it reflects the understanding that whoever controls the layer of local agents controls access to the planet's most sensitive data: that which resides on personal and corporate computers.
From Cloud to Device Core
The story of enterprise artificial intelligence has followed a predictable pattern for years: models live on remote servers, data uploads to the cloud, and the user receives a response. This model has a natural ceiling in corporate contexts because it involves ceding control of information. Manus disrupts this logic by reversing the flow.
With "My Computer," the agent does not ask you to send your files. It enters your file system, executes scripts, automates locally installed software, and operates within an isolated environment that includes a mechanism for human authorization for each terminal command. This detail is not trivial: Meta understood that the speed of adoption in professional settings depends directly on the user never feeling they have lost control. Each sensitive action requires explicit confirmation. It’s an architecture of trust, not just functionality.
This transition from the cloud to the local device represents precisely the phase of De-Monetization within the technological maturation cycle. What once required a development team, specialized software licenses, and hours of technical work, Manus transforms into a sequence of instructions in natural language. The marginal cost of automating a complex task approaches zero for any user with an active subscription. This is not an incremental improvement over the chatbots of 2023; it is the destruction of the business model for entire layers of process automation services.
Manus has processed over 147 trillion tokens and generated more than 80 million virtual computers since its initial release. These numbers are not ornamental. They indicate that the infrastructure is already proven at industrial scale before the desktop application even reaches the general market. Meta is not betting that this will work; it is scaling something that has already proven to function.
The Paradox of WhatsApp and Its Strategic Implications
There is a detail in Manus's expansion that deserves attention as it exposes a genuine strategic tension within Meta. When the platform began integrating into messaging apps in early 2026, it did so first in Telegram, not in WhatsApp. Meta owns WhatsApp. The sequence seems counterintuitive until one reads the context: the European Commission had indicated that Meta blocked third-party AI assistants' access to WhatsApp.
This means that the most logical expansion for Manus, one that would have generated more immediate distribution, was being hindered by regulatory pressure in the European market. The launch in Telegram was not a strategic preference; it was a tactical response to a real restriction. Manus CEO Xiao Hong described the acquisition by Meta as the opportunity to build on "a stronger and more sustainable foundation without changing how Manus operates or how decisions are made." This statement takes on a new meaning when considering that Manus continues to sell subscriptions through its own channels while gradually integrating into Meta's products.
What Meta is building is not an isolated product. It is a layer of agency that connects its social networks, messaging apps, and now users' operating systems. Each point of integration is a node of data and influence. The agent that organizes your invoices today is the same that can manage your calendar, professional communication, and corporate workflows tomorrow. Monetization does not come solely from subscriptions; it comes from positioning to offer high-margin business services based on a user base that has already delegated operational tasks to the system.
The most frequently mentioned competitor in this context is OpenClaw, acquired by OpenAI just before Manus's desktop release. OpenClaw offers similar agency capabilities in chat but with a significantly more complex technical setup for the average user. Manus counters with QR code access, two levels of model (1.6 Max for deep reasoning, 1.6 Lite for speed), and an interface that does not require prior technical knowledge. The battle is not being won in laboratories; it is being won in the adoption friction.
The Local Agent as a Vector for Professional Democratization
The dominant narrative surrounding artificial intelligence agents tends to focus on operational efficiency, on how many hours are saved, how many processes are automated. This reading is correct but incomplete because it ignores the underlying power shift.
For decades, the ability to automate complex workflows was reserved for organizations with budgets to hire developers, integrate APIs, or acquire enterprise software solutions. A small accounting firm, a freelance consultant, or a three-person startup could not access that layer of automation without prohibitive costs. Manus, by bringing a locally controlled agent to any device with a subscription, is compressing that gap.
A freelancer today using "My Computer" to process hundreds of batch invoices, build no-code automation tools, and manage files autonomously is operating with a capability that required a team 18 months ago. That is Democratization in its most direct form: it is not that technology is cheaper, but that access to the kind of capability that once required corporate scale no longer depends on scale.
However, the risk is equally tangible. An agent with local access to the operating system that is not well-defined becomes a vector for error or data loss with consequences that no cloud chatbot interface could generate. The human authorization mechanism that Manus incorporated is the structural response to that risk, but its effectiveness depends on users employing it judically and not turning it into a bureaucratic step they approve without reading. Augmented Intelligence works when the human remains the arbiter of the important decisions, not when they delegate that responsibility along with the task.
The market for local agents is in the transition phase between Disappointment and Disruption within the technology adoption cycle. Expectations have outstripped what products could deliver for months; now concrete delivery is catching up to those expectations. Whoever sets the standard for trust and simplicity in this phase will not be selling software; they will be defining how the next two generations of professionals work. Technology that empowers the individual without replacing their judgement is not a philosophical promise; it is the only architecture that generates sustained long-term adoption.
