The creation of the Agentic AI Foundation (AAIF) This marks a turning point in the development of agent-based artificial intelligence. Under the umbrella of the Linux Foundation, this new entity is born with the goal of bringing order to a rapidly growing ecosystem that needs common rules, cross-platform compatibility, and open collaboration frameworks.
With the advent of Agentic AI, capable of acting autonomouslyConnecting to real-world tools and coordinating multiple tasks independently, the sector is beginning to look beyond simple chats with assistants. The AAIF aims to be the platform where standards are defined that will enable these agents to operate reliably, securely, and predictably in businesses, public administrations, and technology projects worldwide, including in Spain and the European Union.
The foundation starts with three key technological pillars, donated by its original promoters: the Model Context Protocol (MCP) from Anthropic, the agent framework goose developed by Block and the standard AGENTS.md Promoted by OpenAI. Although these tools could already be used freely, they are now under neutral governance, where any organization can contribute to their evolution without depending on a single provider.
La The Linux Foundation contributes its experience in managing large open-source projects such as Linux, Kubernetes, Node.js, or PyTorch. With the AAIF, it extends that model to agentic AI infrastructure, offering legal, technical, and organizational support to maintain open standards and prevent the technology from becoming locked into proprietary solutions.
Among the founding members and main sponsors are names such as AWS, Google, Microsoft, OpenAI, Anthropic, Block, Bloomberg and CloudflareAlong with dozens of companies in the cloud, finance, e-commerce, and enterprise software sectors, the support of these players reinforces the idea that the standardization of AI agents will be a central issue in the coming years.
A foundation to organize the new wave of agentic AI
Agentic AI describes systems that do not simply answer questions, but They plan actions, coordinate services, and execute complex workflows. autonomously. They can book a trip, manage invoices, interact with corporate APIs, or negotiate with other agents on behalf of a user or a company.
This change of focus, from chat to programs that act for usThis is generating a new market for tools, protocols, and frameworks. However, without clear standards, each vendor could define its own rules and formats, creating a fragmented and difficult-to-integrate environment—a particularly problematic situation for European companies that need to comply with strict regulations regarding security, data protection, and traceability.
The Agentic AI Foundation was created precisely to prevent this fragmentation. Under the coordination of the Linux Foundation, the AAIF is conceived as a neutral space for open governance where companies, developer communities and institutions can agree on technical specifications, best practices and interoperability policies for AI agents.
Figures like Jim Zemlin, head of the Linux Foundation, have emphasized that AI is entering a phase where systems are no longer just conversational assistants, but autonomous agents that collaborate with each otherIn that context, founding the AAIF is a way to ensure that the technological base is built with transparency, continuity and community participation, in the style of what happened with the web itself thanks to open standards.
For Europe and Spain, accustomed to promoting regulations such as the AI Act or the General Data Protection RegulationHaving a foundation that promotes open standards can facilitate the alignment between innovation and regulatory requirements. An ecosystem of interoperable and auditable agents is better suited to the demands for compliance and technological sovereignty being debated today in Brussels and other European capitals.
MCP: the protocol that connects AI agents with data and tools
At the heart of this initiative lies the Model Context Protocol (MCP)MCP, a specification born at Anthropic and released as open source, has quickly become established as a common method for AI models to connect to data sources, APIs, applications, and external tools, both in corporate environments and in solutions geared towards end users.
According to data shared by Anthropic and the Linux Foundation, there are already more than 10.000 servers MCPs in operation, deployed in contexts ranging from developer utilities to mission-critical systems of large companies, including Fortune 500 firms. Many of these implementations run on reference cloud infrastructures such as AWS, Google Cloud, or Microsoft Azure.
The protocol has been integrated into very popular AI platforms, such as Claude, Cursor, Microsoft Copilot, Gemini EnterpriseVS Code and ChatGPT, among others. Its appeal lies in offering a standardized way to expose capabilities to agents, with integrated security and permission control mechanisms, so that organizations can decide with considerable precision which resources each agent can access.
Product managers like Mike Krieger from Anthropic have explained that MCP initially emerged to solve internal integration problems and, over time, became a kind of “lingua franca” to connect AI agents with the real worldBy now moving into the sphere of the Agentic AI Foundation, the protocol gains an additional guarantee of neutrality: it no longer depends exclusively on the roadmap of a single company.
Industry voices, such as Swami Sivasubramanian, head of Agentic AI at Amazon Web Services, insist that placing MCP in an open foundation gives confidence to European and global developers. Knowing that it is a standard is reassuring. open, interoperable and community governed It encourages long-term adoption, which is key for companies that don't want to be tied to a specific supplier.
Goose: a local-first framework for secure workflows
The second major technical block entering the AAIF is gooseBlock is an agent framework developed by Block and released as an open-source project. Its design is geared towards an approach local-firstThat is, it prioritizes the ability to execute and orchestrate agents in controlled environments and close to the datawithout always depending on remote services.
Goose combines language models, an extensible tooling system, and native MCP integration, allowing it to manage complex and repeatable workflowsThe idea is to offer a kind of "orchestration layer" upon which companies can design agents, such as role-based AI agentsthat automate internal tasks, integrate legacy systems, or meet specific security and regulatory compliance requirements.
Block, known for services like Square, Cash App, and TIDAL, argues that Goose provides the necessary infrastructure for agentic AI to advance responsibly. Instead of letting each agent act independently, the framework imposes structures, limits, and policies that help ensure more predictable behavior aligned with business needs.
Manik Surtani, head of open source at Block, has pointed out that we are at a point where AI could become the biggest driver of economic growth since the advent of the internet. In his view, the choice is between a scenario dominated by closed solutions or a ecosystem based on open standards and protocols that allows large and small companies to participate on equal terms.
For European companies with significant regulatory obligations—for example, in the financial or healthcare sectors—goose's approach can be particularly interesting. A framework designed for run agents on local or hybrid infrastructures It fits better with the need to keep certain sensitive data under direct control, something highly valued by entities in Spain and other EU countries.
AGENTS.md: Clear and consistent instructions for agents
The third foundational pillar is AGENTS.mda simple standard driven by OpenAI that relies on conventions such as markdownIts goal is to provide AI agents working with code with a single and consistent source of project-specific instructionsso that they know how to behave within a given repository.
In practice, AGENTS.md acts as a "user guide" for the agents: it describes which parts of the code are prioritized, which commands can be executed, which build systems are used, and which style rules must be followed. Thanks to this structure, the Agent behavior is more predictable and errors are reduced when interacting with complex projects or multiple technologies.
According to the data shared, more than 60.000 open source projects and frameworks Many developers have already adopted AGENTS.md, including well-known development tools such as Amp, Codex, Cursor, Devin, Factory, Gemini CLI, GitHub Copilot, Jules, and VS Code itself. This widespread adoption reinforces its role as the de facto standard in the world of agents for programmers.
OpenAI, in addition to promoting AGENTS.md, has also been one of the first to incorporate MCP and has contributed to the ecosystem with tools such as Codex CLI, Agents SDK y Apps SDKAlthough the company maintains closed business models, its commitment to open standards in the agent layer aims to make it easier for organizations to integrate agent AI without being tied to a single vendor.
From a European perspective, having specifications like AGENTS.md facilitates the integration of agents in organizations that already work with distributed repositories, CI/CD tools, and international teams. By having a uniform and machine-readable conventionThis reduces friction when adopting agents within established development flows.
Massive support from the technology and financial industries
The Agentic AI Foundation isn't starting alone. In addition to Anthropic, OpenAI, and Block, the list of members includes... Platinum It includes Amazon Web Services, Bloomberg, Cloudflare, Google, and Microsoft, forming a support bloc that spans from cloud computing to financial data and content delivery networks.
Along with them, numerous companies are joining in the categories Gold and SilverCompanies like Cisco, IBM, SAP, Salesforce, Shopify, Snowflake, Docker, Ericsson, Adyen, Twilio, SUSE, Uber, and Zapier, among many others, are involved. This mix of enterprise software, payment, observability, and cloud service providers indicates that interest in agent standardization extends beyond the major AI firms.
From the financial side, Bloomberg has described MCP as a fundamental building block for APIs In the era of agentic AI, the company sees this protocol as a way to connect data systems, internal tools, and complex analysis workflows, so that professionals can make decisions supported by agents that always have access to up-to-date and relevant information.
Cloudflare, for its part, highlights that open standards like MCP allow for the creation of a more dynamic developer ecosystemwhere anyone can build and deploy agents on different platforms without fear of being "locked" into a single infrastructure. The company has collaborated closely with Anthropic to provide remote MCP support and has seen a significant increase in the number of servers deployed on its network.
In the open source sphere, Google highlights its history of promoting community projects and emphasizes that the mass adoption of new technologies depends on shared standards that developers trustMicrosoft expresses itself in similar terms, pointing out that the future of agentic AI will only be sustainable if it is built with transparent and collaborative paths.
Open source, global competition and Europe's role
The creation of the AAIF takes place in a context in which the debate on the opening up of AI It's very much alive. Many US companies monetize their models through closed APIs, while in parallel, open model initiatives have emerged and multiple agents, both in the West and, especially, in China, with players like DeepSeek, Alibaba, Moonshot AI or Z.ai gaining ground among developers and researchers.
Some analysts warn that the advancement of open models and tools in Asia could translate into a long-term strategic advantageIf other geographical areas do not promote equally accessible alternatives, a foundation like the AAIF, linked to the Linux Foundation, can become a benchmark for Europe, which seeks to balance innovation, competition, and the protection of fundamental rights.
The European Union has already put forward frameworks such as the AI Actwhich aims to establish transparency, security, and accountability requirements for the deployment of AI systems. Having open standards for agents—such as MCP or AGENTS.md—makes it easier for European companies to audit, document, and control behavior of these systems, something especially important in regulated sectors such as banking, health, energy or public administration.
For Spain, where many SMEs and large corporations are beginning to explore the use of AI agents in areas such as customer management, process automation, and e-commerce, having a standardized and neutral infrastructure It reduces barriers to entry. It allows you to test different solutions without being locked into a single provider and increases the likelihood of developing your own solutions based on these technologies.
Furthermore, the combination of open efforts from companies like OpenAI—with initiatives such as gpt-oss or command-line tools for developers—and the competitive pressure from open models in other countries could incentivize European universities, research centers and startups to build upon these standards and contribute their own extensions, adapted to local needs.
AAIF Events, Technical Community and Next Steps
To consolidate this ecosystem, the Agentic AI Foundation will integrate existing community initiatives. A prime example is the Dev Summit, an event and set of podcasts focused on the MCP ecosystem, which now fall under the AAIF umbrella thanks to the contribution of the company Obot.ai.
The next MCP Dev Summit will be held in New York in April 2026The event, with an open call for proposals, attendee registration, and sponsorships managed through the Linux Foundation's usual channels, is designed as a meeting point for developers, system architects, and product managers working with MCP and related tools.
Also planned is a Dev Summit Europe 2026The venue and dates for this event will be announced later. This initiative aims to create a global community of AI agents, with a particular focus on the European market, where discussions about interoperability, ethics, and regulation carry significant weight.
Those who wish to learn more about the work of the Agentic AI Foundation can visit the AAIF.io portal and the official repository on GitHubwhere technical documentation, protocol specifications, reference examples, and training resources will be published. The intention is that both large corporations and small projects can participate in the design and improvement of standards.
With the addition of AAIF to its portfolio of initiatives—which already included Linux, Kubernetes, Node.js, OpenSSF, PyTorch, and RISC-V—the Linux Foundation consolidates its position as a nerve center for open standards in software, hardware, and now, agentic AI infrastructure. If the venture is successful, AI agents operating in Spanish, European, and other companies worldwide could share a common language, facilitating integrations, reducing costs, and fostering more balanced competition between large and small providers.
