Every year, the industry descends on Barcelona with a collective sense of possibility. The stands get bigger. The messaging gets bolder. And this year, the word on every screen, every banner, and every keynote slide was the same: AI. The ambition on display was genuine. So was the activity. What was harder to find, when you looked closely, was AI operating at scale – embedded in production, driving measurable outcomes, running reliably inside real enterprise workflows. The gap between pilot and scale is where the industry is stuck, and MWC 2026 made that gap very visible.
AI at Scale: Still the Exception, Not the Rule
I went in looking for examples of generative and agentic AI that had moved beyond the pilot stage: deployed at scale, integrated into live operations, delivering measurable value. There were bright spots, but they were outnumbered by something else entirely. The harder I looked, the more I found the same pattern repeating: sophisticated marketing wrapped around technology that, when interrogated, was still operating in controlled conditions rather than at production scale.
One stand offered what looked like a compelling example: a telecom agent integrated into a mobile app, designed to help customers navigate their plans. I asked it a straightforward question. Given my profile, what is the most optimised plan for me? The response was generic, with no contextual information, no personalisation, no retrieval. The kind of output you would expect from a pilot that had not yet been connected to live customer data. Promising as a proof of concept. A long way from scale.
Across the floor, the pattern held. AI was present in almost every conversation, and in many cases the underlying intent was serious. But production-grade deployment, the kind that runs reliably inside complex enterprise environments and delivers outcomes at volume, remained the exception. The distance between what was being demonstrated and what was actually running in operations was, in most cases, considerable.
The Conversations Happening on the Ground
Away from the stands, in meeting rooms and over coffee, a different conversation was happening. Senior contacts across major operators were wrestling with questions that were almost entirely operational: how do I bring my systems together? How do I manage my data? How do I reduce the cost of my application landscape and make better use of legacy investment? These are not the questions of an industry on the cusp of an AI revolution. They are the age-old questions that have defined telco transformation for a decade, and the fact that they remain front of mind tells you something important. The foundation required for meaningful AI deployment is not yet in place for most operators. Nobody has greenfield. Any agentic capability built on fragmented, unmanaged data will collapse under the weight of its own promises.
Why the Impasse Persists – and How to Break It
Part of the problem is structural. The AI ecosystem today is effectively a layered stack: infrastructure providers and foundation model builders at one end, enterprise applications and last-mile connectivity at the other. What is missing in the middle is the capacity to bridge the two – to take powerful, generic technology and make it work within a specific enterprise context, against real data, inside real operational constraints. Without that, the gap between what AI can do in a demo and what it can do in production remains vast.
The more productive conversations at MWC were the ones that started from context rather than capability. Not which AI platform should we buy, but: given what we have and what we need, what is actually worth attempting? That shift in framing opens up more interesting territory. Rather than searching for new software to replace traditional systems, why not use AI to rewrite them, preserving the stable business logic whilst re-engineering everything around it? Rather than debating whether an agentic contact centre is theoretically possible, why not build one in parallel and find out empirically which workloads can be transferred and which cannot? These are not visionary questions. They are practical ones. And they tend to produce answers – particularly when you are working with an AI Builder who understands both the enterprise context and the frontier technology well enough to know what is genuinely feasible.
Scaled AI in telecoms is not theoretical. We have delivered it. Working with a major UK operator, we have implemented large language models at scale within customer operations and service assurance workflows, using SaaS-based capabilities adapted and productised into live business processes. The work is not a pilot. It is running in production, handling real volume, and demonstrating that the gap between AI ambition and AI delivery is closeable, with the right approach and the right partner.
Two Paths Worth Taking
None of this means agentic AI has no place in the near-term telco roadmap. It means the approach has to be honest about where the industry actually is. Two pragmatic paths emerged from my conversations at the show.
The first is building agentic capability in parallel with existing operations. For a contact centre, this means keeping your current workforce and processes intact while standing up a parallel capability to handle defined tasks: password resets, simple troubleshooting, root-cause triage, call diversion. You experiment, measure, and transfer incrementally. You do not bet the organisation on technology that has not yet earned that level of trust.
The second is building a thin agentic layer on top of existing legacy rather than replacing it. The underlying workflows and data remain in place. The agentic layer calls on that data intelligently, executes against defined parameters, and handles interactions that previously required human intervention. Neither path is a revolution. Both are how real progress actually gets made.
The Dimension Nobody Is Talking About
There is a third element to this challenge that received almost no airtime at MWC, and yet may be the most consequential of all: people. The best technology fails without human trust, readiness, and behavioural change. You can deploy the most sophisticated agentic layer imaginable, but if the teams working alongside it do not understand when to trust it, when to override it, and how to improve it, the investment will not return what you need it to.
This requires clear human-in-the-loop design: defining where the AI acts autonomously, where humans approve or guide outcomes, and how the system learns from every intervention. It requires visible escalation paths and feedback loops that demonstrate to people their input is actually shaping the model. And it requires targeted capability building: AI literacy for business users, supervisory skills for managers, prompt and model fluency for technical teams. One-size programmes serve none of these groups. The operators who move fastest will treat AI competency as a core performance expectation, and will find ways to recognise the people who make human-AI collaboration work in practice.
What Will Separate the Winners
The operators who benefit most from this transition will not necessarily be those with the biggest budgets or the most aggressive AI strategies on paper. They will be the ones willing to experiment with parallel models, build incrementally on top of legacy, invest seriously in the human side of adoption, and resist the pressure to change the entire ocean at once. Progress will come from those who are clear-eyed about where they are starting from, disciplined about where the value actually sits, and honest enough to ignore the noise. MWC 2026 showcased extraordinary ambition. The use cases that justify that ambition, at meaningful scale, remain elusive. That is not a counsel of despair. It is an invitation to have a more honest conversation about what transformation actually looks like from where telcos are standing today. If that is a conversation you want to have, we are ready for it.
Shashi Bhagavathula is a senior leader at Cognizant with deep expertise in the telecommunications sector. Pavan Malladi is a senior leader at Cognizant specialising in commercialising technology across the Communications, Media and Technology sector. To explore how Cognizant can support your transformation journey, get in touch.
