For Kenyan banks, telcos, and large enterprises, the next phase of AI may not be about adding another chatbot to their websites. The bigger challenge could be connecting AI agents to the systems that already run their businesses.

That was the key message WSO2 delivered at WSO2Con Africa 2026, where the company argued that businesses cannot deploy AI agents across their operations without first connecting those agents to their existing systems.

An AI agent can be built to carry out tasks on its own, but it is only as useful as the information and systems it can actually reach.

It’s relevant for large Kenyan organizations in particular, since most of them run separate systems for things like managing customers, processing payments, handling compliance, running HR, and keeping internal operations moving.

For a bank, for example, an agent handling a customer request may need information from several systems before it can complete a task. A telco could similarly have customer, billing, network, and service systems that need to work together before an agent can take action.

WSO2 argues that many existing enterprise systems are designed primarily for interaction with other software or access by developers rather than autonomous AI agents.

This means businesses looking to introduce agents may first need to make their existing infrastructure “agent-ready.”

WSO2’s Integration Platform is designed to provide this layer. The company says its WSO2 Integrator product can connect enterprise systems and workflows, while its platform has more than 600 connectors that can be used to connect applications and services.

The company is also using Model Context Protocol (MCP) to expose enterprise data and functionality to AI agents. MCP servers can act as an interface through which an agent interacts with existing systems.

For Kenyan businesses, adopting AI does not require tearing out the software they already use. Instead, they can connect their AI agents to those existing systems, letting the agents handle tasks that pull information from several departments or applications at once.

This moves them beyond simply answering questions and allows them to help with actual business processes.

WSO2 also highlighted long-running workflows, which allow a process to stop when it needs information or a decision from someone, then continue once that step is completed.

In a loan application, for example, an agent could handle some of the checks, pause while waiting for a customer to provide information or an employee to approve the application, and then continue with the remaining steps.

The approach also raises questions around control.

WSO2 says its platform provides human-in-the-loop guardrails that can limit what an agent is authorized to do. An organization could, for instance, allow an agent to handle certain transactions automatically while requiring human approval for actions above a defined threshold.

The company also highlighted tools for evaluating agents during development and monitoring their actions in production.

For Kenyan enterprises, therefore, the AI challenge extends beyond choosing an AI model or building an agent.

The organizations that want agents to perform real business functions will also need to consider how those agents access data, interact with existing applications, and operate within established controls.

That makes integration less of a back-end technical concern and more of a central consideration in how enterprises introduce AI into everyday operations.