AI in government is moving from policy discussions and experiments into practical public services across Africa, but technology leaders from Uganda, Zambia and Oman say the difficult work often begins before an AI model is deployed.
At WSO2Con Africa 2026 in Nairobi, the panel on the role of AI in government focused on a familiar set of constraints: fragmented data, integration, infrastructure, skills and the need to define the public problem before choosing the technology. The discussion brought a practical dimension to the wider conference debate about how governments can move from digital systems toward services capable of using AI to act on behalf of citizens.
The panel featured Tonny Bbosa, Manager of Business Transformation at Uganda’s National Information Technology Authority, Harrison Chapu, Director of Innovation at the Zambia Revenue Authority, Hamda Al Issaei, Integration Developer at Oman’s Social Protection Fund, and Dev Wijewardane, Senior Director and Principal Enterprise Architect at WSO2. WSO2Con Africa is taking place in Nairobi from September 22 to 24 under the theme “Building the Agentic Enterprise,” with government AI forming part of its broader discussion about integration, identity, data and AI-enabled operations.
AI adoption starts with the problem, not the technology
For Chapu, the starting point for a government AI project is the same as any other technology programme: identify the problem first. He said organisations need to establish whether AI is actually appropriate for the problem, because conventional automation can sometimes provide the required solution without introducing an AI system.
That assessment also has to consider whether the institution has the necessary skills and data. Once a use case has been approved, the implementation follows the broader technology lifecycle, including development, testing, external user acceptance, deployment and post-implementation support. Government services add another layer because the people affected by a system are often outside the organisation, making external stakeholder feedback part of the process rather than an optional final step.
Bbosa made a similar argument from Uganda, where he said technology selection begins with defining the problem and establishing the intended outcome. With public budgets constrained, he said his team looks at measurable changes in service delivery, such as reducing the number of physical visits a citizen needs to make or shortening a process that previously took several days.
That approach also provides a way to distinguish an AI project that produces public value from one that simply demonstrates a new technology. If a government service becomes faster, requires fewer physical interactions or can be delivered remotely, the institution has a measurable basis for assessing the result. NITA-U’s existing digital-service work illustrates this broader direction: its eBiz platform brings services from multiple government entities into a single online interface, while its UGhub platform has been designed to support data sharing across government systems.
Government data is becoming the main constraint
The panel repeatedly returned to data because digital transformation has already produced large quantities of information inside government. The challenge is whether that information is accurate, sufficiently structured, interoperable and appropriate for the particular AI application being considered.
Chapu described the situation at the Zambia Revenue Authority as a mixture of accumulated data and continuing limitations. The authority has internal and customs data as well as information obtained through interfaces with other government institutions, including the ministries responsible for lands, agriculture and finance. The data, however, remains segmented, while legacy systems create questions around data attributes and integrity.
That distinction becomes particularly important when governments move from advisory AI toward systems that can make or execute decisions. Chapu said imperfect data does not necessarily prevent governments from deploying AI for advisory applications, where a human remains responsible for the decision. The threshold becomes different when an agent is expected to make consequential decisions without a person in the middle.
His conclusion was direct: the AI itself is not necessarily the central problem. The platform underneath it has to provide the data and systems required to make the technology reliable enough for the intended use.
Al Issaei made the same connection from an integration perspective. She argued that differences between identifiers, missing standards, poor-quality data and weak organisational integration can all be reflected in an AI solution. Her position was that governments need sound integration and data before they can expect AI to solve problems built into those underlying systems.
The point is particularly relevant as governments consider AI agents that can access several systems and take actions across them. An agent operating across agencies inherits the quality, permissions and interoperability of the systems it can reach. Better models cannot, by themselves, resolve inconsistent identifiers or establish authority over data that has never been connected.
AI services are already producing practical results
The discussion did not suggest that African governments are waiting for every digital foundation to be perfect before using AI. Al Issaei pointed to a diabetic retinopathy screening project involving more than 35 healthcare institutions, where AI is used to analyse retinal images and support the identification of possible problems.
She was clear that she was not part of the project, but used it as an example of how an AI application can translate into a direct public-service benefit. At the same time, she identified data preparation, integration and security as practical challenges involved in taking such systems into production.
Uganda is also building on infrastructure that can support wider digital services. NITA-U describes UGhub as an integration platform intended to enable government entities to share data securely and efficiently, with capabilities including API management and identity and access management. The authority’s stated objective is to improve service delivery as well as government reporting, planning and decision-making.
That infrastructure gives AI applications somewhere to operate. It also illustrates why government AI cannot be considered separately from the digital systems that have been built before it.
Integration comes before more ambitious AI agents
The progression described by the panel is important. Governments have spent years digitising individual processes, accumulating data and connecting selected systems. The next challenge is making those systems work together reliably enough for AI to use them.
This is closely related to the wider agentic architecture discussed at WSO2Con Africa. The conference agenda describes AI agents as systems that interact with existing enterprise capabilities, while other sessions focus on APIs, integration, identity, governed access and AI-ready data.
For government, however, the implications are broader because an agent may operate across institutions rather than within a single organisation. A citizen service could eventually require information from several government databases, a transaction with another agency and an approval or escalation by a human official.
Bbosa sees that development as part of Uganda’s longer digital-government trajectory. He pointed to the country’s existing data-exchange infrastructure as a foundation for applying AI to government processes, with the potential to reduce physical visits and administrative costs. NITA-U’s own description of UGhub similarly places data sharing across government entities at the centre of its service-delivery architecture.
That makes interoperability a practical prerequisite for more ambitious agentic government. An AI agent cannot coordinate a citizen journey across agencies if the agencies cannot reliably exchange the information and services that journey requires.
Civil servants need to understand AI before deploying it
Technology infrastructure is only one part of the preparation. Chapu argued that governments also need to address a basic knowledge gap among civil servants, beginning with a clearer distinction between conventional automation and artificial intelligence.
He said employees need to understand what AI can do, what it cannot do and where ordinary process automation remains sufficient. That foundation matters because employees who do not understand the technology may either expect too much from it or treat it as a threat to their roles without understanding how their work will actually change.
The skills requirement will also vary by role. Technical teams may need to build or manage AI systems, while legal, business and operational teams require enough understanding to use those systems responsibly within their respective functions. Chapu argued that training therefore needs to be structured around the different levels of AI use inside an organisation rather than treating every civil servant as though they require the same technical expertise.
This is consistent with the wider public-sector challenge around AI governance. Once an AI system is embedded in a public service, questions about accountability, data access, security and human intervention become part of the operating model rather than issues that can be left to a technical team.
The next government model could become more proactive
The panel’s closing discussion moved beyond today’s AI applications toward a government that can anticipate needs rather than waiting for citizens to initiate every interaction.
Bbosa described a future in which citizens would need to make fewer physical visits to government offices, while services could become faster and operating costs could fall. He also associated greater use of digital systems with more transparent and predictable government processes.
Al Issaei described a similar direction in terms of proactive services, where government systems can use available information to understand a citizen’s circumstances and provide relevant services without requiring the citizen to navigate government structures themselves.
Chapu offered a more specific example from revenue administration, describing a future in which information shared between government institutions could allow a revenue authority to identify economic activity before the taxpayer reports it directly. He noted that an interface with Zambia’s Ministry of Finance already exists, although he said it was not yet being used for the particular scenario he described. The example was therefore a projection of what connected government data could enable rather than a description of an existing AI service.
That distinction matters because proactive government creates a different set of questions from an AI assistant answering a citizen’s request. Once government systems can infer circumstances from information held elsewhere and initiate actions, questions about authorisation, data sharing, transparency and accountability become central.
The broader lesson from the panel is that African governments do not need to wait for every underlying system to become perfect before experimenting with AI. They do need to match the sophistication of an AI application to the quality of its data, the maturity of its integration layer, the skills available inside the institution and the consequences of getting a decision wrong.
The path toward agentic government therefore runs through work that can appear less visible than an AI interface: connecting systems, improving data quality, establishing standards, defining permissions, training public servants and measuring whether services actually improve. As governments take those foundations further, AI can move from isolated applications toward systems capable of coordinating more of the work that sits behind public services.
Real ESG impact doesn’t happen in panels alone, it happens in the rooms where financiers, operators, and policymakers actually align. Our GreenShift Forum 2026 cuts the noise, bringing together the people rewiring Africa’s sustainability and energy frameworks for one focused day in Nairobi. Secure your seat.
Go to TECHTRENDSKE.co.ke for more tech and business news from the African continent and across the world.
Follow us on WhatsApp, Telegram, Twitter, and Facebook, or subscribe to our weekly newsletter to ensure you don’t miss out on any future updates. Send tips to info@techtrendsmedia.co.ke
