From AI adoption to the AI-native telco: Five takeaways from the AI-Native Telco Forum

The AI-Native Telco Forum in Düsseldorf came at a pivotal point in the industry’s AI journey. After years of pilots with generative AI, automation and increasingly capable agents, operators now face the harder question: 

How can AI innovations move safely and economically into production at scale?

That changes the debate. Proving that a model can answer a question or an agent can complete a task is not enough. AI must work inside the telco itself, interacting with customers, networks, operational systems and critical infrastructure. Adoption has become a question of operations governance and delivering business value.

1. AI is now a business and investment decision

A recurring message in Düsseldorf was that boards are not interested in the number of AI use cases deployed. They want to know what those investments are delivering: revenue, cost reduction, customer experience, capacity, efficiency and workforce impact. The emphasis has shifted from adoption to measurable business benefit.

That requires a more economically grounded approach. It calls for the right workload, compute and outcome. Not every use case needs the most powerful model. Some decisions must happen at the edge; others can be handled centrally. Start with the outcome, speed, information, cost and value created.

2. The industry has an execution gap

The gap between experimentation and scaled operation runs through several dimensions of the telco. Legacy systems remain a constraint, but there is no single answer: modernize around them, add agentic layers, or replace fragmented estates over time. Phased transformation may be realistic, but intelligent agents cannot mask inconsistent processes and data.

Scaling AI solutions is not solely an AI problem. It is an architecture, data, operating-model and organizational-readiness problem. Leadership, culture, skills and organizational design matter as much as technology. Operators must decide how work gets organized when people, software and autonomous agents are all involved in delivering outcomes.

3. Autonomy must be bounded, trusted and observable

Agents were central to many conversations in Düsseldorf, but full autonomy is neither immediately achievable nor entirely desirable. The key question is where autonomy creates a better outcome and what degree of autonomy is appropriate. Recommending a next best action is very different from changing a network configuration, approving a transaction or acting directly on a customer account.

That points towards bounded, trusted autonomy: clarity on which decisions an agent can make, what information it can access and when human approval is required. It also avoids a simplistic “human in the loop” model that turns every AI decision into another manual queue.

Trust also depends on how quickly operators can verify what AI has done. If an agent acts in milliseconds but the organization confirms the result much later, speed of execution has outpaced speed of control. Operators need to know what an agent did, why it acted, whether it stayed within policy and how to intervene.

4. Data, architecture and standards determine what AI can achieve

Trusted data, shared definitions, fewer silos, real-time information and accessible context are fundamental to scaling AI. It is not possible to make dependable real-time decisions from incomplete, inconsistent or delayed information. If the information arrives hours later, the architecture has already limited the outcome.

Composability matters for the same reason. APIs, events, modular components and data products affect how quickly the organization can sense, decide and act. The industry must also avoid a new generation of agentic silos that recreate today’s interoperability problem in a more intelligent form.

That is why AI standards are important now. They must go beyond interfaces to cover agent identity, permitted behavior, shared semantics, observability, governance and agent-to-agent interaction across technology and organizational boundaries.

5. A shift from isolated use cases to end-to-end outcomes is required

A portfolio of useful AI implementations can generate local improvements without changing overall business performance. A customer-care agent may respond faster, a network agent may diagnose faults more quickly and an engineering assistant may improve developer productivity. Each provides value, but the outcome may still be constrained by the hand-offs between them.

The more meaningful unit of transformation is the end-to-end outcome: preventing churn, resolving an issue before it becomes a complaint, dynamically creating and assuring a network service, or bringing a multi-party proposition to market faster. This is where business-critical scenarios and autonomous workflows across IT, network and operations are essential. They define the outcome first, then the capabilities, data, processes, KPIs and reusable ODA assets needed to deliver it.

The sequence is deliberately business-led: define the outcome, map the value streams, identify the capabilities, apply reusable ODA assets, govern execution and measure the result. It creates a common line of sight across business, IT and networks, giving agents a governed environment in which to act.

That changes transformation from disconnected projects into continuous capability building. The AI-native telco will not be defined by how many models or agents it has deployed, but by whether it can sense, decide, act across business, IT and network domains, verify the result and learn from it. That is the shift from AI adoption to an AI-based operating model built around value.

Watch the panel discussion at the AI-Native Telco Forum in full: Closing the AI attribution gap between deployment and financial outcomes | TelecomTV