The collision course: traditional APIs meet AI-native integration

While telcos celebrate the GSMA’s OpenGateway initiative (https://www.gsma.com/solutions-and-impact/gsma-open-gateway/) as their digital renaissance, a fundamental shift in machine-to-machine communication threatens to render it irrelevant before it fully launches.

The culprit? Model Context Protocol (MCP).

There are many conversations, articles and posts claiming we are facing a technological inflection point that telco executives need to confront immediately. As AI agents increasingly mediate software interactions, the entire premise of manually-integrated APIs is being challenged.

The question isn’t whether telcos should offer APIs. The question is whether anyone will be still using them in five years the same way we are doing today.

OpenGateway might have the right vision, but does it embrace the right implementation?

OpenGateway promised to break down walls between telecom capabilities and software developers. Standardized APIs across operators. Unified access to network slicing, edge computing, and quality of service controls.

The vision sounds compelling, but the timing may be catastrophic. That’s because while telcos have been meticulously designing their API strategies, AI development has fundamentally changed how software interacts with systems.

MCP: The communication protocol built for AI, not humans

The API paradigm have long been the standard for software integration. It was built for deterministic software and human developers who write explicit integration code. The model, simply put, is pretty straightforward: a developer makes a request, an API returns a structured data, and an application processes it. This paradigm worked, and works, perfectly when software is deterministic (the data structure is known) and rule-based. Its limitations become glaring when autonomous AI agents enter the picture:

• Static endpoints vs. dynamic interactions: APIs require predefined request formats and response handlers. AI agents need to discover, interpret, and adapt in real-time.
• Explicit integration vs. implicit understanding: Each API requires custom integration code. AI agents need to understand capabilities inherently.
• Isolated contexts vs. persistent understanding: Traditional APIs don’t maintain context across services. AI needs continuous contextual awareness.

This is precisely the gap that MCP aims to fill.
Model Context Protocol reimagines machine interactions from first principles. Instead of rigid request-response patterns, MCP creates an AI-native layer where systems expose their capabilities through natural language descriptions, interactive documentation, and contextual examples.

This, for instance, enables AI agents to:

• Dynamically discover available capabilities without developers writing custom integration code
• Maintain context across multiple services and providers without friction
• Interpret and adapt to system responses rather than requiring explicit error handling

The key insight: When software is mediated by foundation models, communication between systems must be optimized for those models, instead of doing so for human developers.

So, are Telcos and OpenGateway in risk of falling behind?

Right now, OpenGateway is fundamentally API-driven. It’s built around the idea that developers will integrate telecom APIs into their applications. But what happens when most developers stop writing API calls and start deploying AI agents that interact autonomously?

Telcos could be left behind if AI-driven applications move beyond API-based interactions.
Cloud providers and AI-first companies could take over by offering AI-native integration layers that are more flexible and efficient.
Telecom infrastructure could become commoditized, with telcos reduced to “dumb pipes” rather than active enablers of AI-driven connectivity.

The fact is that we have see similar disruption before. Cloud computing abstracted infrastructure management away from traditional data centers. Core and Radio Access Networks have been softwarized, abstracting the functionalities from the underlaying infrastructure. MCP could do the same with API.

The question is: Are telcos already considering MCP? The answer is… unclear.

Some forward-thinking operators are experimenting with AI-driven network management and intent-based networking, which share conceptual similarities with MCP.
OpenGateway itself has not publicly addressed MCP, but if AI-driven development continues at its current pace, ignoring this shift could be a strategic mistake.
Companies like OpenAI, Anthropic, and Google DeepMind are rapidly developing AI agents that interact dynamically across different systems. If MCP or a similar paradigm takes hold, telcos that fail to adapt may lose developer engagement to cloud-native competitors.

Telcos don’t need to abandon OpenGateway, but they should evolve it behind its current form.
Continue API development while simultaneously investing in MCP compatibility layers. Don’t force an either/or decision when the market remains in flux. Launch targeted pilots where AI agents can directly interact with network capabilities through natural language understanding rather than rigid API calls. Fostering the cross-industry collaboration by engaging with AI companies developing agent frameworks to ensure telco capabilities are represented in emerging standards rather than being bypassed.

Shifting from API-first to AI-first development isn’t speculative: every major cloud provider ins investing in AI agents that abstracts away explicit API calls. Telecommunication providers don’t shine for being ahead of the market. Historically, telcos preferred arriving a bit late to a technological transition. The future of initiatives like OpenGateway will also depend on whether Telcos will recognize the and embrace the shift before cloud and AI companies take the decision for them.