Telco AI agents face trust test

Vendors like Wavelo and Amdocs are pioneering agentic AI deployment through real-time, event-driven architectures and AI-driven automation integrated across comprehensive OSS/BSS platforms. Wavelo’s platform streams operational and network events continuously, enabling AI agents to autonomously respond to issues such as network congestion while remaining governed by business policies and human oversight, as Bob Dietrich emphasized the importance of 'governed collaboration between specialized agents.' Meanwhile, Amdocs leverages its extensive portfolio to enable telecom operators to achieve real-time autonomous decision-making and service delivery, marking a transformative shift from traditional rule-based systems to AI that acts with intent and adapts continuously to network conditions.

Complementing these real-time capabilities, other vendors like RADCOM Ltd. and Circles Australia are addressing foundational challenges such as telco data silos and customer experience integration to make networks truly AI-ready. RADCOM’s cloud-native platform tackles data quality issues head-on, enabling seamless AI operations, while Circles Australia’s full-stack 'core to edge' platform delivers unified operator portals and super apps that enhance customer retention through dynamic monetization strategies like cashback rewards. This holistic approach underscores the industry’s move toward integrated, cloud-native AI-driven OSS/BSS ecosystems that unify multiple telecom applications into customer-centric experiences.

Innovations extend into AI orchestration platforms that blend cutting-edge AI models with deterministic execution to ensure reliability and prevent hallucinations. Kaya Global Inc., for example, employs large language models during design and testing phases to enable prompt-based configuration of agentic workflows but enforces deterministic runtime execution to maintain operational integrity. This hybrid approach balances AI creativity with the telecom industry's stringent demands for predictable, trustworthy automation.

Practical demonstrations of agentic AI’s potential have been showcased through TM Forum Catalyst projects, where Wavelo’s Telemetry and Billing Agents dynamically orchestrated network performance monitoring and monetization workflows during large-scale events. These projects illustrate how autonomous AI agents can deliver tangible business value by enabling real-time assurance and commercial automation within complex IoT ecosystems, reinforcing the case for incremental adoption strategies that integrate AI agents gradually into existing OSS/BSS frameworks with human oversight, as advocated by Wavelo’s leadership.

Telecom operators confront a profound trust deficit in deploying agentic AI, as only 14% can substantiate claims of AI trustworthiness despite 72% expressing confidence in their systems. This skepticism stems from concerns over reliability, bias, transparency, and accountability, especially when AI agents operate autonomously without human oversight. Microsoft, Apple, Cisco, and Salesforce have collectively emphasized that robust governance, identity, and security controls are now non-negotiable prerequisites for enterprise AI deployment, underscoring the critical need for comprehensive governance frameworks to manage risk and build operator confidence.

The integration of agentic AI within telecom environments is complicated by legacy infrastructure and siloed data, which impede seamless interoperability and elevate operational risks. Operators face challenges in ensuring data quality, compliance, and secure authorization across complex workflows, as highlighted by SAP’s insistence on ERP systems as the trusted system of record for governed enterprise data. Vendors are innovating with digital twins and super apps to bridge these gaps, yet the human-in-the-loop dilemma remains unresolved, necessitating clear policies and architectural designs that balance automation with human oversight to mitigate irreversible or high-risk actions.

Operationalizing agentic AI demands a paradigm shift in governance infrastructure, moving from experimental pilots to production-grade observability and control. Microsoft Agent 365 exemplifies this evolution by positioning itself as a control plane for monitoring, securing, and governing AI agents, reflecting the new baseline expectations for telecom operators. Additionally, frameworks that embed legacy operational governance—such as control gates evaluating security, SLA impact, and rollback capabilities—ensure AI-driven network changes are safe and accountable, with human approval retained as the final safeguard against unintended consequences.

The complexity of telecom networks, with thousands of interdependent components, challenges AI’s ability to replicate expert human judgment, necessitating domain-specific models and digital twins to guide AI decision-making. This complexity also reshapes operator roles, shifting staff from direct network management to supervising and retraining AI agents in an 'expert-agent collaboration' model. Such human oversight is essential to align AI actions with organizational priorities and risk thresholds, preventing misaligned decisions and maintaining trust in agentic AI systems.

Huawei’s introduction of the A2A-T protocol marks a pivotal step toward establishing a unified communication standard designed to coordinate thousands of autonomous AI agents across telecom network layers and systems. Positioned as a foundational control plane akin to TCP/IP for the Internet or Kubernetes for cloud computing, A2A-T enables efficient interconnection by providing a universal language for agents to exchange intent, goals, and state information while incorporating mechanisms for negotiation and conflict resolution to mitigate risks of emergent unintended behaviors. This protocol exemplifies the broader industry push for layered intelligence and open collaboration, aiming to mask integration complexities across multiple network generations and vendors, thereby allowing communication service providers (CSPs) to focus on global resource orchestration and long-term policy iteration.

The layered intelligence approach advocated by Huawei and embraced by over 100 industry partners and approximately 30 CSPs facilitates real-time autonomous closed loops within individual domains while enabling global collaborative orchestration across domains. This dual-layered strategy reduces delayed responses and integration complexity by isolating single-domain intelligence from cross-domain orchestration, effectively accelerating the deployment of highly autonomous networks. As Eric Yang, President of Huawei Carrier Business, highlights, such open collaboration and unified standards like A2A-T are critical to scaling agentic AI operations that can autonomously optimize network performance, customer experience, and service assurance at unprecedented scale.

The telecom industry’s evolution toward supporting not only human users but potentially hundreds of billions of intelligent software agents underscores the urgent need for standardized multi-agent AI coordination protocols. Huawei projects that these autonomous agents will act on behalf of people, businesses, and devices, vastly expanding the addressable market and intensifying demands on network intelligence and operational agility. By defining the architecture for agent communication and championing open collaboration, Huawei’s A2A-T protocol aims to create a scalable framework that can harmonize diverse vendor platforms and generations, thereby enabling the agentic operations that could redefine growth trajectories for telecom operators worldwide.

Enterprises are rapidly moving beyond isolated AI pilots toward full-scale agentic AI deployments that leverage multiple specialized AI agents collaborating to automate complex workflows. For example, a Southeast Asian logistics firm cut vendor onboarding from five days to just four hours by chaining four AI agents responsible for contract extraction, ERP validation, compliance checks, and executive reporting, achieving 99.8% accuracy and freeing procurement staff to focus on strategic negotiations. This transition is largely driven by board-level mandates demanding measurable bottom-line impact, prompting technology leaders to embed AI structurally into core processes with rigorous feasibility studies and phased rollouts targeting high-impact, low-risk use cases.

The evolution from standalone chatbots to multiplayer AI agents acting as 'digital teammates' is transforming enterprise collaboration by enabling AI agents to participate in shared workflows alongside human teams. Gabriel Hubert of Dust highlights how users can '@' mention specific agents—such as a blog writer passing content to a LinkedIn agent—facilitating seamless task handoffs that enhance cross-functional cooperation. In sales, for instance, AI agents automate lead qualification and CRM updates in workflows that continuously improve and become accessible across the team, standardizing operations and reducing process times.

As multiplayer AI agents gain access to sensitive enterprise data, robust governance and access controls become critical to prevent unauthorized exposure. Dust’s approach, as explained by Hubert, ensures that AI agents inherit the data access permissions of their respective collaborative spaces, maintaining consistent security regardless of user. This shift also necessitates new organizational roles like 'AI operators,' who holistically rethink workflows by questioning whether existing processes remain relevant in an AI-enabled environment, effectively running an 'anti-to-do list' to eliminate redundant tasks and optimize enterprise efficiency.