Agentic AI promises smarter, more autonomous telecom operations, but fragmented OSS/BSS, middleware, and messaging platforms deny it the operational context it needs. Without end-to-end visibility, CSPs risk weaker automation, poorer customer outcomes, and missed AI-driven revenue opportunities, writes meshIQ.
The telecom industry is booming. Analysts expect that the market for operational and business support systems (OSS and BSS) will exceed $54 billion by 2031. The U.S. has over 99 percent 5G penetration, and global cellular IoT connections grew 16 percent year-over-year in 2024. Yet, all of these operational environments rely on event streams, messaging brokers, APIs, and integration layers – platforms that currently operate in isolated environments and provide fragmented oversight for communication service providers (CSPs).
This separation between applications is problematic on its own, but the rise of agentic AI exposes how costly fragmented operations can become. Without end-to-end visibility, AI systems struggle to deliver reliable automation, recommendations and operational insights.
Disconnected platforms frequently prevent agentic models from accurately recommending improvements, triggering workflows, or automating repetitive tasks. The result is not only weaker automation but also increased risk to customer experience, operational efficiency and revenue-generating processes. While 54 percent of CSP executives predict AI will significantly contribute to their revenue in the next three years, the current environment makes effective integration difficult.
For agentic technology to scale across CSPs, executives will need to connect their fragmented systems and create cross-platform visibility across all technology layers. This starts by understanding why these intelligent models can’t function across independent architectures.
When intelligence lacks visibility
Agentic AI needs operational context to work correctly in telecom and mobile/wireless industries. Sixty-seven percent of CSP executives cited integrating and managing complex data, and 52 percent cited legacy IT infrastructure as top AI implementation challenges. This means operators are struggling to apply intelligent technology to systems that can’t communicate with one another. They lack the connectivity needed to reflect what’s actually happening across the entire business.
An AI agent can’t operate correctly on a generic infrastructure signal. It needs to understand all subscriber activations, device registrations, billing transactions, IoT data streams, and customer app integrations to provide insights that a CSP executive can use.
Let’s consider a hypothetical scenario where a telecom operator is trying to use AI to predict subscriber activation failures before the service issues reach customers. This technology would need to correlate activity across billing platforms, customer-facing applications, messaging technologies, APIs, and more. If, for example, messages accumulate in a Kafka topic, the downstream provisioning workflow may receive information too late to complete the transaction.
With platforms running in isolation, autonomous models would struggle to pinpoint exactly where the issue occurred and could generate unreliable predictions, leaving CSP executives with insufficient information to act. In practice, this can translate into delayed activations, increased support tickets and customer dissatisfaction despite all systems appearing healthy in isolation.
Faced with these context gaps, some operators consider expanding the volume of data available to AI. This logic makes sense, as most would assume that more data is a reasonable solution for context issues. But this is not always the case.
Drowning in telemetry
Many operators assume that increased telemetry volumes can bridge context gaps and provide AI with the information it needs to operate effectively. In reality, this creates more operational noise than the technology can realistically deal with. The challenge is not a lack of data. It is the inability to connect data to business transactions and operational workflows across multiple platforms.
Wireless networks generate large volumes of data approaching petabyte scale, which can help AI draw on a broader base of information, but this is not a replacement for connected workflows. Without a unified middleware layer, these networks could produce vast amounts of network alarms, device registrations, and IoT sensor data that automation would struggle to decipher. These signals don’t provide AI with a more complete understanding – they just make it more challenging for intelligent systems to determine which signals are meaningful and require evaluation, and which are low-value, everyday clutter.
Additionally, these telemetry volumes have exceeded what operators can analyze manually, forcing organizations to rely on AI to understand patterns across their business transactions. Therefore, carriers require a complete picture of enterprise infrastructure not only to make agent-based models work correctly, but also to interpret the growing volumes of network data available to the industry.
To start fixing these issues, operators need to veer away from data-centered improvements and focus on the foundation that supports AI at scale: middleware.
Preparing telcos for agentic AI
To allow AI to work effectively and analyze large volumes of telemetry, service providers often follow a common progression.
1 | Unify middleware management and observability. Carriers should aim to bring streaming, legacy messaging, and hybrid cloud infrastructures under a single umbrella view. This creates standardized observability into data flows between OSS/BSS platforms, 5G architecture, IoT ecosystems, and open RAN networks.
2 | Correlate events across business transactions. Once middleware environments are visible, operators should connect messaging activity to the business processes it supports, such as subscriber activation, billing, provisioning and IoT workflows. This helps AI distinguish between isolated technical alerts and issues that may affect customers or revenue.
3 | Apply AI to the base. The prior steps focus on unification and building the operational context AI needs to function as intended. The last step focuses on integrating AI into this established base. In practice, these intelligent models should provide more reliable recommendations, automate repetitive workflows, and detect discrepancies, helping operators move from reactive defenses to predictive operations.
While this framework is not comprehensive, and there are many actions an enterprise could take to improve its middleware operations, these steps should give CSP executives a starting point for effective AI implementation.
The path forward for telcos
Telecom and mobile industry professionals are dealing with increasing data volumes and event streams across 5G, IoT, and open RAN networks. This telemetry could greatly expand the information AI can access and act on, but only with a unified middleware environment that gives that information meaning.
If organizations want agentic AI integrations to work effectively, they should move beyond legacy middleware approaches. Unifying middleware and messaging platforms can provide automation with the foundation it needs to work effectively. When done correctly, this consolidation provides AI with the operational context needed to move from reactive troubleshooting to predictive operations. For CSPs, middleware visibility may ultimately determine whether agentic AI becomes a transformative capability or simply another isolated technology initiative.
