At Telco Days, the Test for Agentic AI Will Be Control, Not Conversation
By Sam Ellis · Technology & Health Correspondent
Software Mind will hold its online Telco Days 2026 conference on 6 November 2026, bringing together a familiar set of telecom priorities: agentic AI, network operations, fraud prevention and the commercial systems that sit behind customer service.
The agenda matters less as another AI event than as a practical indication of where the industry’s attention is moving. The question is not simply whether an AI model can summarise an incident, draft a response or recognise a suspicious message. It is whether an operator can allow software to take, recommend or coordinate actions in systems where a mistake can affect billing, connectivity, fraud exposure and customer trust.
According to Software Mind’s announcement, the free event will run from 9.00am to 5.30pm CET and feature contributors from Telefónica, Amdocs, Alianza, Next Mobile, ROCCO, Tallence and IBM. A ROCCO session is due to present findings from more than 300 telecom leaders on AI adoption, while other sessions will address charging operations, voice AI and SMS fraud. These are claims about the planned programme rather than evidence of outcomes from the event itself. (softwaremind.com)
The operational context is the difficult part
Agentic AI refers broadly to systems that can pursue a task through several steps, using tools and information sources within predefined limits. In telecoms, that could mean helping to investigate a network fault, validating an enterprise order, identifying a potentially fraudulent transaction or routing an issue to the appropriate team.
That potential makes the technology more consequential than a general-purpose workplace assistant. Telecom workflows often cross network, billing, product, care and security systems. They are also highly rules-driven, with legacy platforms and data held in separate places. An answer that sounds plausible is not sufficient if the next step changes a configuration, blocks a message, delays a payment or gives a customer the wrong explanation.
The value of telecom AI depends on whether its actions remain intelligible, reversible and accountable.
This is where the event’s emphasis on production use cases is more useful than broad claims about AI maturity. PwC’s 2026 analysis argues that telecom operators are active experimenters but lag leading organisations in the practices that turn projects into repeatable enterprise value: trusted data, reusable components, redesigned workflows, accountable owners and regular portfolio decisions. It recommends using agents in bounded, high-value workflows with explicit guardrails and measurable outcomes. (pwc.com)
For an operator, those outcomes should be specific. A network-assurance tool might be assessed against incident resolution time and unnecessary engineer call-outs. A care workflow might be judged by repeat contacts, escalation quality and complaint levels. A fraud intervention should be measured not only by suspicious activity stopped, but also by false positives and the impact on legitimate customers.
Security is also a customer-experience issue
The Telco Days agenda includes a session on generative AI’s role on both sides of the SMS channel: supporting fraud attempts and helping to detect them. That framing is credible. Mobile messaging remains a route into phishing, impersonation and account compromise, while generative tools can make fraudulent content quicker to create and more convincing at scale.
The GSMA has warned that generative AI is contributing to the rise of scam-as-a-service models, and says mobile operators are working on fraud and security controls intended to protect both customers and network integrity. Separate GSMA work on scam-bot detection also points to infrastructure-level detection of abnormal signalling patterns as a way to identify SIM-farm activity before scams reach consumers. (gsma.com)
This is why security cannot sit apart from service design. Over-aggressive filtering, poorly explained account checks or a weak appeal route can shift the inconvenience of prevention on to legitimate users. Conversely, a fraud response that recognises suspicious behaviour quickly, gives customers clear information and preserves access where possible can protect confidence in the operator’s service.
A programme about evidence, not attendance
Software Mind says its event will address the gap between AI experimentation and measurable results. The test for the sessions on 6 November is therefore straightforward: do they distinguish a promising demonstration from a workflow that has clear ownership, reliable context, effective safeguards and an agreed commercial or customer measure?
The broader market context suggests that distinction is overdue. GSMA Intelligence reports that recent operator AI activity has increasingly included enterprise propositions, sovereign infrastructure and revenue objectives, alongside longstanding efficiency ambitions. The opportunity is real, but it will be decided through operational discipline rather than presentation language. (gsmaintelligence.com)
For telecom leaders, the most valuable lesson may be modest: start with decisions that can be observed, challenged and improved. If an agent cannot explain the information it used, stay within a defined authority or demonstrate a better outcome for the customer and the business, it has not yet earned a place close to a live network.
This article is based on a supplied announcement. It has been edited by Expert View Media for clarity, context and length.
Comments