How a Global Telecom Provider Built an Omnichannel Support System Without Rewriting Its Agents

Customers do not think in terms of product silos. They think in terms of problems.

A customer might start a conversation about mobile connectivity, continue it through a smart TV interface, and escalate it through a call center - expecting the system to understand context, history, and intent across all channels. In practice, each handoff resets the conversation. Context is lost, memory is fragmented, and customers are forced to repeat themselves.

The root cause is architectural, not conversational. Each support agent was designed to be effective within its own product boundary. None were designed to collaborate with other agents, share long-lived context, or reason jointly about a customer’s broader situation.

Attempting to stitch these systems together through point-to-point integrations or shared databases quickly led to brittle dependencies and inconsistent behavior.

For a provider operating at telecom scale, this approach simply did not hold.

A critical constraint shaped the solution from the start. The existing agents could not be replaced.

They were already deployed in production, tightly coupled to their respective products, and built using different technologies. Rewriting them into a single framework or migrating them into a unified environment would have taken years and introduced unacceptable risk.

What was needed was not a new “super agent,” but a way for existing agents to collaborate without losing their autonomy.

Instead of centralizing logic, the team reframed the problem. Each product-specific support agent would remain responsible for its own domain. The omnichannel experience would emerge from coordination, not consolidation.

This required a shared layer that could sit beneath all agents and provide three critical capabilities: discovery, shared context, and memory propagation.

Enter BAND.

BAND was introduced as an agentic mesh that connects independently deployed agents across environments, frameworks, and organizational boundaries. Rather than acting as a new support agent itself, BAND provides the infrastructure that allows existing agents to behave as part of a single system.

Each support agent registers with the mesh and declares its capabilities, such as which products it supports and which customer interactions it can handle. Agents remain deployed where they are - in their original clouds, clusters, or on premise environments - but become discoverable and addressable through the mesh.

When a customer interaction begins in one channel, the handling agent publishes context and memory updates into the mesh. If the conversation moves to another channel or product, the next agent does not start from scratch. It retrieves the shared context, understands prior interactions, and continues the conversation seamlessly.

From the customer’s perspective, the system behaves like a single omnichannel assistant. Under the hood, it is a distributed collaboration between many specialized agents.

A key design principle was avoiding a centralized “conversation brain.”

No single agent owns the full conversation or dictates the flow. Instead, context and memory are treated as shared system resources, propagated through the mesh. Each agent contributes what it knows and consumes what it needs.

This approach avoids creating a new bottleneck or single point of failure while allowing agents to remain focused on what they do best. A smart TV support agent does not need to understand mobile billing logic, but it does need to understand that the customer has already been authenticated, that a network issue was identified earlier, or that an escalation is already in progress.

BAND enables exactly this form of selective context sharing.

In an omnichannel support system, failures are rarely isolated. A breakdown in one channel often surfaces in another.

By routing agent-to-agent communication through the mesh, the provider gained end-to-end observability across the entire support journey. It became possible to trace how a customer interaction moved between agents, which context was shared, where decisions were made, and why a particular escalation occurred.

This level of visibility was critical not only for debugging, but also for compliance, quality assurance, and continuous improvement of the support experience.

The introduction of an agentic mesh fundamentally changed how omnichannel support could evolve.

New support agents could be added for new products or channels without redesigning the entire system. Existing agents could be upgraded or replaced independently. Most importantly, customer experience improved without forcing organizational or technological convergence across teams.

Engineering teams remained owners of their respective agents, while the platform team focused on shared infrastructure and governance.

Omnichannel support does not fail because conversational AI is immature. It fails because organizations try to solve a distributed systems problem with monolithic designs.

When agents are treated as isolated applications, omnichannel behavior is impossible. When they are treated as participants in a shared system, it becomes achievable - even across clouds, stacks, and organizational boundaries.

This telecom provider did not unify its agents. It unified their ability to collaborate.

BAND was the mesh that made that possible.