At the very end of 2025 (12/29), Meta announced the acquisition of Manus, an Artificial Intelligence (AI) startup based in Singapore but founded in China. The deal is estimated to be worth between 2 and 3 billion dollars. Initially, many interpreted the news as “just another move in the Artificial Intelligence race.” A comfortable reading, but an incorrect one.
Meta did not merely buy an AI startup. It acquired operational capability and is internalizing a critical capacity because it understood something fundamental: the value of AI is not in the insight, it is in the execution.
But why is that?
The year 2026 will mark the moment we definitively enter the era of agents: systems that go beyond chatbots, as they answer questions, execute tasks, make assisted or unassisted decisions, interact with corporate systems, and generate real business impact. The chatbot is the new email, the agent is the new digital collaborator.
And this will change how companies operate. Over the last two years, the market has been obsessively focused on the question of which is the best model: GPT, Claude, Gemini, Llama?
The acquisition of Manus makes it clear that this is no longer the right question. The correct question has shifted to: who controls the orchestration, execution, governance, and integration of these models with data and processes within the real context of businesses?
Models are becoming accelerating commodities, and companies are seeking value not in better model responses, but in faster decisions, automated processes, reliable agents operating within clear rules and with governance, controlled costs, and compliance from day one.
At our company, this vision is not new; it is foundational. We created a platform (MATH AI Platform) precisely as a result of understanding that the future of corporate AI would not be mono-model, mono-cloud, or mono-use case.
Our strategy is based on clear principles: AI needs to be orchestrated, not just consumed; agents must operate with fallback, governance, and explainability; corporate data is the center of intelligence, not the model; and the right model is the one that makes sense for each context, cost, and risk.
While the market debated “which LLM to choose,” we were building fallback between multiple providers and models, domain-specialized agents, deep integration with data, APIs, and processes, and layers of LLMOps, FinOps, and governance. In other words, an infrastructure for the age of agents.
The message Meta sent (even without saying it) is that the purchase of Manus is a clear signal that conversational AI is merely a user interface, but the true value will lie in agents, which will act autonomously to enhance company operations. .
Those who do not master orchestration and governance will become hostages to vendors and may pay a price later.
Big Tech has already understood this. Companies that still treat AI as a feature or a Proof of Concept (POC) are behind. Leaders don't need to know how to use it, but who is in control of their AI when it starts to act.
For us, in our projects, AI only generates value when it is governed, integrated, and execution-oriented.
Thiago de Morais Dutra, Executive Director of Research and Development at MATH Group.
