Andy Linham of Vodafone Business talks about prioritising use cases, starting slow to go fast safely, building the AI Booster, MCP, A2A and the missing agent discovery protocol…
In conversation with Mobile Europe‘s Editor, Annie Turner, Vodafone Business’ Principal Strategy Manager, Andy Linham began by explaining his is a central role within the Vodafone Group. His team’s job is to figure out things that will help its many local operating companies and their customers.
The interview took place in The Briefing recently on Separating telcos’ AI milestones from moonshots – in other words recognising what could deliver tangible business results as opposed to a massive solution to a huge problem using disruptive technology. One does not necessarily exclude the other – it’s all about figuring out the most appropriate approach, which depends on the organisation’s goals.
The group identified more than 900 potential AI use cases then devised a framework to prioritise the most useful on two axes – with feasibility as the vertical and access/benefit on the horizontal. Each use case is ranked.
WATCH THIS INTERVIEW NOW ON PLAYBACK
Customers’ acceptance is key
According to Linham, some examples which were judged to be not immediately technically feasible but had huge benefits for customers were prioritised. He noted, “Coming from a technology company, it’s really easy to start with the technology and say, ‘Oh look, I’ve got this new cool thing that can do this. What can I then do with it?’
“What we spend a lot of time doing is talking to our customers about what they’re interested in, what they would accept from an AI perspective. That’s part of the reason why, if you look at some of our network developments as an example, we haven’t prioritised automated AI creation of networking capabilities because customers didn’t want us to.”
He added, “Our job isn’t to be leading the charge, saying, ‘AI is the greatest thing ever. It’s going to change your world’. Our job is to be helping our customers to fundamentally understand what it is and how it works…As soon as they understand how these [models] work and that they are effectively glorified statistical calculators, they become…more trusting…They’re now more thinking, ‘OK, if the risk isn’t there, how can I use it?’.”
Building the AI Booster
Vodafone’s next step was to build the AI Booster. Perhaps surprisingly, Linham said, “Being a telco, we are fortunate enough to deal in a very heavily regulated environment…I genuinely think that helps us operate as a better company…we have a slight different lens to lots of other industries.”
And this is one reason the AI Booster is key: requests are routed through the Booster which checks for things like prompt injection attacks, or any kind of security threats or anything not condoned by the company or by the EU AI Act, for example. If the request is approved, it passes through to the AI service and the requester receives their response.
Linham said Vodafone’s early focus on security and regulation and protection probably meant the operator was slightly slower with some initial use cases, now it can go faster “because we have that level of protection for what we’re doing internally”.
He stresses that guardrails and regulation are an enabler and the Booster has a team that constantly updates it and looks at integrating new models.
Ideas skyrocket from AI-ready data
Vodafone has been something of a trailblazer regarding its approach to managing data (there’s a good summary in this interview), even so when it began its relationship with Google Cloud a decade ago, nobody was thinking about how to apply AI beyond the analytics kind. Linham outlined how tough it has been getting the operator’s data to where it can work with AI and says, “We’re now at a really good place”.
Vodafone has a history of importing innovation from local markets for export to the wider group and Linham highlighted innovations from Vodafone Turkey which has super strict data privacy regulation.” That too speeded progress. Linham said, “Once you’ve got the data – that is by far the hardest bit from my perspective – once it’s in a single location, or…at least within a documented set of locations, all of a sudden the opportunities just skyrocket from there.”
Standards impact integration
He said integration was less of an issue because Vodafone has long insisted on standardised tools and TM Forum for standardised north-south APIs and Mplify Alliance for east-west services. “that…has meant that when it comes to integrating different tools with different parts of data with different AI services, we’ve made sure that that kind of standards layer is also built into AI Booster.
“So if I want to call a certain system, I can talk to AI booster using that same sort of integration method. It just sits as another layer now in our orchestration framework, and that means that the actual use of AI and that data has been less painful than it would have been had we started from scratch.”
He gave more detail about how Vodafone uses various large language models, which are governed by various parameters including cost.
MCP, A2A…faster, better, smaller – and bigger
When asked about if he’d known what he knows now at the beginning what he’d have done differently he says the biggest surprise has been the relentless page of change. For example, that within six months or a year the Model Context Protocol (MCP) would appear which allows an AI solution to communicate with an application and extract data from it.
Then a few months later the agent to agent (A2A) protocol hits the market so one agent can talk to another independently of human involvement “and these thing just keep snowballing and the models keep getting better, they keep getting faster, they keep getting smaller although in some cases they keep getting bigger. It means you’re constantly having to kind of rebalance the types of things you are developing and building.
“And that’s it’s brilliant, right? It’s so exciting to see all these things coming really, really quickly. But originally, when it first started happening, I had this mindset that right, this is now the future. We’re now here…I’ve learnt my lesson now. I’m not taking anything for granted.”
Standards, de facto standards and missing links
The good thing is that despite the constant changes, Linham said it’s a case of case of building on and adapting what Vodafone has done so far rather than fundamental change or retro-fitting. This is because of a standard-based approach although he acknowledges MCP and A2A are not official standards, approved by a recognised standards body, but they are in wide use.
“It means that if we want to change a supplier out because we found either a better one or a more cost-effective one, or the one we were working with doesn’t have the right solution anymore, as long as the replacement supplier supports MCP and has that kind of integration, we’re we’re quite prepared to swap them out,” he said.
However, one thing that’s holding up progress, according to Linham, is, “There’s no standard for agent discovery at the moment… [that] if I need to find an agent to do a certain thing, this is how I find it. This is how I authenticate it. This is how I authorise it. Those are the bits that we’re still kind of working around in a more manual process at the moment.”
There’s a lot of activity in this area. For example, Infoblox is looking at things called DNS A that repurposes the approach behind the internet’s domain name system (DNS) which converts web pages into IP addresses. The Cisco Outshift team have done work called Agency (AGNTCY) which is backed Google, Dell, Red Hat and others.
Linham concluded, “The joy is because of the way we’ve got this modular architecture, if we do have to change how we integrate from one set of tools to another set of tools, we’ve typically got…building blocks in between [that] allow us to change one component piece of the architecture because that’s where the integration happens. So it’s not going to be a huge amount of work for us.”
