Building the future: Advances in Ai infrastructure for autonomous agents

The views expressed are those of the author at the time of writing. Other teams may hold different views and make different investment decisions. The value of your investment may become worth more or less than at the time of original investment. While any third-party data used is considered reliable, its accuracy is not guaranteed. For professional, institutional or accredited investors only.

The evolution of AI infrastructure for autonomous agents is reminiscent of the transformative impact that Transmission Control Protocols (TCPs) and other foundational protocols had on the early internet. Just as TCP/IP provided the essential framework for reliable data transmission and communication across diverse networks, modern AI infrastructure is laying the groundwork for autonomous agents to operate seamlessly and efficiently in complex environments. These advancements are not only enhancing the capabilities of individual agents but also enabling a new era of interconnected, intelligent systems that can collaborate and adapt in real time, much like the internet revolutionized global connectivity and information exchange.

Emerging protocols

For example, San Francisco-based Anthropic’s Model Context Protocol (MCP) marks a pivotal step forward, enhancing decision making and autonomy in AI applications, making them more reliable and versatile. Google also recently introduced its new A2A protocol, which standardizes interoperability across frameworks and vendors. A2A fosters a dynamic ecosystem in which AI agents securely exchange information and coordinate actions, enhancing productivity and creativity.

While protocols like HTTP, TCP, and POP3 are established communication protocols, standardization has allowed significant infrastructure to be built around them. We believe the world is at a Cambrian moment that will see new protocols unleash seismic shifts in infrastructure build. According to Grand View Research, the AI platform market is poised to hit US$1.8 trillion by 2030, growing at a compound annual growth rate (CAGR) of 35.9%. Generative AI, a key driver of this expansion, is projected to have a CAGR of nearly 50% over the same period.¹ Amidst the market expansion in large language model (LLM) technology, the autonomy is likely to define the next phase of AI advancements.

Challenges in agent development

Delivering autonomy in the form of agents presents challenges, however, and those challenges are prompting the emergence of new protocols. This brings us back to MCP. MCP is essentially a standard connection that provides a universal extension point for LLMs and development tools to connect with databases, ticketing systems, and more. Initially designed to enhance productivity of integrated developer environments (IDEs) like Anthropic’s Claude Desktop, Claude Code, VS Code, Cursor, and Windsurf, the concept is now gaining popularity in other areas. MCP servers are among the fastest-growing code repositories on GitHub, a collaborative developer platform, underscoring the rapid adoption and development in this corner of the AI sector.

Need for robust data infrastructure

As useful as they are, MCP servers still require infrastructure to support their development and tooling to manage their implementation. Current MCP implementations appear to have significant vulnerabilities, however. These make it easy for attackers to access secure shell (SSH) keys and other private credentials. MCP servers alone cannot fully solve the challenges inherent in agent development and management — including assessing data quality, connecting LLMs with dynamic enterprise data, data processing, scaling data-volume capacity, and integrating heterogeneous data systems. Poor or insufficient data directly limits an agent's ability to provide accurate insights and make sound decisions.

This confluence of factors — the rise of powerful agents and the limitations of existing data infrastructure — creates a critical need for "agent infrastructure." Traditional data architectures, often characterized by siloed databases, batch-oriented data integration processes, and inconsistent data representations across systems, are ill-equipped to meet the demands of complex AI agent tasks. The rapid proliferation of AI agents is thus creating a demand vacuum for infrastructure capable of reliably serving consistent data across heterogeneous sources.

Today, innovative early-stage companies are aiming to solve these challenges. See a complete list of Wellington’s investments across private markets.