Vodafone and Microsoft call for standardization on how AI agents access skills

An increasing number of telcos are building AI skills libraries to organize and manage how AI systems perform recurring operational tasks. And as their use grows, Vodafone and Microsoft are among the companies that would like to see TM Forum standardize how agents access Open APIs within skill libraries.

AI skills libraries stock standardized, reusable modules that AI agents can draw on to perform multi-step operational tasks. People and agents can request tasks using natural language, doing away with the need for code to tap into technical resources.

In an interview with TM Forum’s Insight, Hitesh Kumar, SVP and Head of Consumer Tech & Digital Innovation describes how IOH’s internally developed AI library lets people use natural language to access a range of skills from architecture and scaling to storage and pipeline management.

“If you have never done coding in your life, you would still be able to do full-stack development using IOH’s skills for front-end, Java back-end, pipeline management, IaaS [infrastructure-as-a-service], event-driven architecture, testing and more,” according to Kumar.

Ease-of-use aside, other benefits of an AI skills library include allowing telcos to define and catalog what agents should do and how.

For example, “IT administrators can ensure agents work in a deterministic manner by creating skills that resemble an eTOM workflow around product lifecycle management,” points out Rick Lievano, CTO, Worldwide Telco and Media Industry, Microsoft.

And IT administrators can update AI skills to ensure that agents comply with the latest governance requirements (although a mis-managed or out-of-date library also has the potential to propagate errors across multiple agents).

In addition, they can help centralize and standardize the capturing and sharing of knowledge, according to Dr. Lester Thomas, Head of New Technologies & Innovation, Vodafone Digital & IT.

“When humans are doing work, it is very hard to have continuous improvement and very hard to share expertise,” says Thomas. “With agent skill libraries you capture institutional knowledge – and measure, evaluate and continually improve every aspect of the skill.”

Making ODA available to agents

The first agent skill Vodafone built for its library was the documentation of requirements for its software teams because “no one likes doing documentation”, says Thomas.

The primary aim was to speed up internal processes. But it also resulted in much higher levels of consistency and quality, according to Thomas.

“Documentation is [now uniformly] structured, so there is a much better result and the agent came up with lots of inconsistencies and missing links,” he explains.

Vodafone, which has direct operations in 17 countries, is using its library of AI skills to “amplify” work on the Open Digital Architecture (ODA) and “make it usable to agents”, Thomas says.

When it comes to architecture, for example, agents can draw on the skills library to capture requirements.

“Engineers working on AI systems use [AI] skills to help build [ODA] components. They don’t build components from scratch,” Thomas explains. “The agent knows how to build it, so it is consistent and better.”

So far, Vodafone has built more than 40 AI agent skills for its software development teams, after starting in September.

Using the skills means “we are making engineers more productive, as well as delivering higher quality and more standards compliant software,” says Thomas.

“We’re setting out to do the whole lifecycle for software delivery with AI,” he adds.

AI agents get ready for business

Whereas Vodafone’s AI skills library is making the lives of its internal developers easier and turbo-charging its use of ODA, Microsoft’s Lievano emphasizes that AI skills give business teams a standardized and reusable approach to using AI agents.

For although technology teams provide the shared governance layer, “it’s the business that is really creating and managing those agents”, says Lievano.

He points out that TM Forum’s Open APIs have a a critical role to play within AI skill libraries because they already standardize interconnection and interoperation across backend systems.

“TM Forum Open APIs are gold”, Lievano explains. “If you’re already standardized on those and your suppliers are using them, then it’s relatively easy for anyone or anything to connect to those APIs consistently, regardless of who the vendor is.”

However, AI agents use model context protocol (MCP) to discover and invoke API capabilities. The lack of a bridge between MCP servers and TM Forum Open APis “is the biggest technical challenge for business users to address”, says Lievano.

He would therefore like to see TM Forum make it simpler for agents to use Open APIs by exposing reference MCP definitions alongside the Open API specifications.

Microsoft has already built an ODA AI gateway, which is a cloud-based component that bridges between MCP servers and any TM Forum Open API. IT teams can use it to create an AI gateway for TM Forum Open APIs and then make them available to business users.

But Lievano would like TM Forum to take this further. Agent skills are portable for use by agents within any AI environment, which means they are ideal for encapsulating standards. Lievano therefore sees an opportunity for the provide definitions and instruction for agents that are common across any platform “and that you can implement in your tool of choice ... be it Copilot Studio, Gemini or Foundry””, Lievano says, adding that this would help even the smallest telcos benefit from AI skill libraries.

“TM Forum can play a critical role in standardizing and simplifying the work agents perform. Much of what an agent needs to do is fundamentally the same from one telecom operator to another, so there’s no reason to reinvent the wheel,” says Lievano.

According to TM Forum’s EVP Products & Services, Andy Tiller, TM Forum members have already made progress with AI-Native ODA. “Our collaboration programs and Innovation Hub projects are packaging ODA’s frameworks and standards as agent-ready knowledge bases and associated agent skills libraries,” says Tiller.

“The AI-Native ODA Roadmap also adds new capabilities to ODA, for example ontologies that give semantic understanding to agents for reasoning,” explains Tiller. “And ODA Canvas extensions that manage the lifecycle of agents, models and data products, and integrate them with existing ODA Components and Open APIs. ODA is evolving from a toolkit used by people to a foundation used by AI agents governed by people.”