Meta acquires Manus: A clear signal of the new level of AI execution

Meta's decision, the parent company of Facebook and Instagram, to acquire Manus for over $2 billion represents one of the clearest signals of how competition in AI is evolving. It's no longer just about improving model quality: there is growing attention to the operational layer that turns AI reasoning into concrete, end-to-end work.

The announcement has been confirmed by both sides and echoed by authoritative sources, in a context where Meta is accelerating AI investments to counter Google, Microsoft, and OpenAI, shifting the focus from conversational demos to systems capable of producing artifacts, managing complex workflows, and operating with minimal supervision.

Manus as an execution engine, not as a chatbot

Manus, founded in Singapore by Chinese entrepreneurs and officially launched in early 2025, has always stood out as a true execution engine. Its general-purpose agent is designed to plan complex tasks, orchestrate tools, iterate on intermediate outputs, and deliver finished results: research, code, analyses, and structured plans, not just textual responses.

Manus was founded in Singapore by Chinese entrepreneurs and officially launched in early 2025

This approach has shown rapid results: Manus has surpassed two million users on the waitlist and has beaten OpenAI agents and other frontier systems in the GAIA benchmark, which measures the ability to complete real multi-step tasks, often with margins above 10%. These are signals not only of reasoning but, above all, reliability in execution.

Usage metrics that speak to production, not experimentation

In the acquisition announcement, Manus stated that it had processed over 147 trillion tokens and created more than 80 million “virtual computers,” key indicators of sustained usage in production contexts. Meta confirmed that the platform will continue to operate from Singapore and offer the service on a subscription basis, with teams and technology integrated into the group’s AI organization.

Co-founder and CEO Xiao Hong, known as “Red,” will report directly to Meta’s COO, Javier Olivan, strengthening the strategic relevance of the deal within the company.

What users were really doing with Manus

One of the most interesting elements that emerged before the acquisition concerns the platform’s actual use. According to the official Discord community, users shared “replayable” sessions documenting the complete execution of tasks: not simple prompts, but structured processes such as climate-change research reports, data-driven visualizations, historical model comparisons, multi-location travel planning with budgets, and dedicated manuals.

Manus was used to synthesize advanced scientific literature, propose directions for academic research, and design self-sufficient homes with precise engineering constraints

On the technical front, Manus was used to synthesize advanced scientific literature, propose directions for academic research, and design self-sufficient homes with precise engineering constraints. And examples that show how the agent operates in the so-called “messy middle” of enterprise AI: too complex for a single prompt, too open for traditional automation.

A development pace that strengthened its positioning

The success of Manus has been accompanied by a particularly aggressive development pace. With Manus 1.5, released in October 2025, the company redesigned the core engine to handle long workflows: average completion times fell from about 15 minutes to under four, thanks to dynamic reasoning allocation and wider context windows.

In December, version 1.6 extended capabilities toward creative and multi-platform workflows, including mobile app development, image generation and editing, presentation creation, and full-stack application building that the agent could test and correct autonomously. An evolution that has solidified Manus as a persistent execution system, not just an input-output interface.

The value lies in orchestration, not in the proprietary model

A key aspect of the operation is that Manus does not train proprietary models. It relies on third-party models, including those from Anthropic and Alibaba, focusing its differentiation on orchestration, reliability engineering, and context management. Yet, according to public sources, Manus had reached about $100 million in annual recurring revenue just eight months after launch.

According to publicly reported figures, Manus had reached around $100 million in annual recurring revenue just eight months after launch

This reinforces a widely held view: durable value does not necessarily reside in foundation models, but in the application layer that makes them useful, interchangeable, and monetizable. In this light, Meta did not acquire a “model company,” but an off-the-shelf agentic infrastructure.

Strategic implications for enterprises and platforms

For enterprise decision-makers, Manus's acquisition is less an endorsement of a single provider and more a strategic signal. Agent orchestration, tool management, retry logic, memory, and auditability become critical assets, equal to the underlying models. At the same time, Meta's track record with enterprise products invites caution: initiatives like Workplace have struggled to maintain a coherent long-term roadmap.

Manus could have an immediate impact in Meta's consumer and small business spaces, where an execution-oriented agent can natively integrate into high-frequency surfaces like Facebook and Instagram, automating content creation, interaction management, advertising, and reporting.

The next defensible layer of the AI stack is the one that turns intelligence into completed work

Ultimately, the deal signals that the next defensible layer of the AI stack is the one that translates intelligence into work actually completed. And it is precisely for this level that today large platforms are willing to invest billions.

For companies, this is a lesson on where to focus investments and talent: not just models, but execution pipelines, context management, reliability, and the ability to integrate tools and workflows end-to-end. In this sense, Manus is not just a startup: it represents a vision of how AI development and use in enterprises could evolve, with a focus on the actual production of useful and measurable work.

Fonte startup-news.it