AI at Work in East Africa: When Jobs Change by Law?

The legal question begins before a job disappears

Public discussion about artificial intelligence often asks how many workers will be replaced. Employment law usually asks a more precise question: what did the employer actually do to the job? A business may introduce AI and keep the same number of employees, but change their duties, performance measures, reporting lines or working hours. It may move some workers into new roles and abolish others. It may use software to score performance or allocate work. Only in some cases does it remove the position completely.

This distinction matters because the International Labour Organization’s 2025 work on generative AI concludes that job transformation is more likely than wholesale replacement for many occupations. East African employers should therefore not wait for a dismissal before asking whether employment law has been triggered. The legal risk often begins at the moment the business redesigns work, collects new employee data, changes performance expectations or decides which roles will survive a technology led restructuring.

AI can turn an efficiency project into a redundancy process

Uganda now provides an unusually direct example. The Employment Act, as amended in 2026, states that an employer may terminate for redundancy where business operations have ceased or where reorganisation of work, introduction of labour saving devices, a change in work pattern or the need for fewer employees justifies the decision. The language is technology neutral, but it fits an AI restructuring almost exactly. The employer does not gain a special exemption because the labour saving device is software rather than a machine on a factory floor.

Kenya reaches the issue through its redundancy procedure. Section 40 of the Employment Act requires notice, attention to fair selection factors such as seniority, skill, ability and reliability, payment of accrued leave, notice or pay in lieu, and severance of at least fifteen days’ pay for each completed year of service. Tanzania’s Employment and Labour Relations Act treats economic, technological, structural or similar needs as operational requirements and requires notice, disclosure and consultation before retrenchment on the reasons, alternatives, selection, timing and severance. The common lesson is that “AI made the role unnecessary” is a business explanation, not a complete legal procedure.

Consultation should happen while alternatives are still real

Consultation loses meaning if management has already switched on the system, removed access, selected the affected workers and prepared termination letters. Tanzania’s retrenchment rules expressly require consultation on measures to avoid or minimise retrenchment. Uganda’s section 80 requires an employer planning to terminate at least ten employees within three months for economic, technological, structural or similar reasons to provide relevant information to union representatives at least four weeks before the first termination, subject to the statutory exception, and to notify the Commissioner thirty days before termination.

These rules do not create a universal East African duty to retrain every worker whenever software changes a task. That would overstate the law. But redeployment and training can become legally important because they are practical alternatives to dismissal. If a worker can continue in a redesigned role after a reasonable period of training, an employer that never considered that option may struggle to explain why termination was necessary in a system that requires genuine consultation or consideration of alternatives. Rwanda adds a useful post dismissal protection: under Article 22 of Law No. 66/2018 regulating labour, a worker dismissed for economic or technical reasons within the previous six months is entitled to reinstatement without competition if the worker meets the profile for a position the employer seeks to fill.

Changing a job is not the same as abolishing it

Suppose an accounting officer previously prepared reconciliations manually and is now expected to supervise an AI tool, investigate exceptions and approve outputs. The job may have changed substantially without disappearing. The employer should then look first at the employment contract, job description, workplace policies, collective agreement and applicable labour law. A change that stays within the reasonable scope of the existing role is different from a unilateral change to essential terms such as pay, status, location or working hours.

This is where careful documentation protects both sides. The business should be able to show the old tasks, the new tasks, the reason for the change, the skills needed and whether the employee was given a realistic opportunity to adapt. An employee should know whether the change is temporary experimentation or a permanent redesign. Calling every change “innovation” creates ambiguity. Calling every change “redundancy” can be equally misleading. The legal classification should follow what happens to the employment relationship in substance.

AI management creates a data protection problem as well

The employment relationship can change even where nobody is dismissed. The International Labour Organization describes algorithmic management as the use of tracked data and other information to organise, assign, monitor, supervise and evaluate work. A delivery company may score routes and acceptance rates. A call centre may analyse tone, pauses or customer ratings. An office may measure response times, keyboard activity or output. These systems can influence promotion, discipline, pay or termination.

Kenya’s Data Protection Act gives a person the right not to be subject to a decision based solely on automated processing, including profiling, where the decision produces legal effects or similarly significant effects, subject to stated exceptions. The Data Protection (General) Regulations require meaningful information about the logic involved, measures to prevent errors, safeguards against discrimination and bias, human intervention, and data protection impact assessment for specified high risk processing. Rwanda’s personal data law similarly protects a person against solely automated decisions with legal or significant consequences and requires impact assessment for high risk uses, including systematic and extensive automated evaluation and new technologies. An employer cannot therefore treat workforce data as legally weightless simply because the software was bought from a third party.

Discrimination can enter through apparently neutral scores

An algorithm may use attendance, historical promotion data, customer ratings or productivity measures that look objective. The problem is that historical data can carry older inequalities, and proxy variables can reproduce them. A performance system that rewards uninterrupted availability may disadvantage workers with protected caregiving responsibilities. A recruitment model trained on a narrow historical workforce may learn patterns that correlate with gender, age, disability or other protected characteristics even if those labels are removed.

East African discrimination rules differ by jurisdiction, but the practical employer duty is similar: do not rely on the vendor’s assurance that the system is “fair.” The organisation using the system should test what data is collected, which outcome the model predicts, whether the measure is actually relevant to the job, how errors are corrected and whether a human can reverse a harmful result. The ILO’s 2025 work on AI in human resource management warns that flawed objectives, biased data and opaque programming can undermine recruitment, pay, scheduling and performance decisions. The legal risk belongs to the employer making the decision, even when the model was designed elsewhere.

A simple record can prevent a complicated dispute

Before a material AI workplace change, an employer should create a written change record. It should explain the business problem, identify the technology, map the affected tasks, record which roles are changed or removed, state the proposed selection criteria, identify consultation obligations, consider redeployment or training, and document any worker data that will be collected. Where automated decisions or systematic monitoring are involved, the record should connect with the organisation’s data protection impact assessment and human review procedure.

This is not paperwork for its own sake. If a dispute reaches a labour officer, regulator, mediator or court, the important question will often be whether the reason and procedure were genuine. A contemporaneous record is more persuasive than a justification written after dismissal. It can also reveal a problem before it becomes litigation: perhaps the job was not truly redundant, the selection criterion is biased, the monitoring is excessive, or a training option is cheaper than termination.

AI should change the evidence, not erase ordinary labour law

East Africa does not need a completely new employment code every time a new AI tool appears. Existing rules on redundancy, operational requirements, consultation, fair procedure, discrimination, data protection and collective termination already reach much of the problem. What AI changes is the evidence. Employers will increasingly need to explain how a technology altered tasks, why fewer workers were required, how scores were produced and why a human decision was fair.

The practical conclusion is that the legal question behind the productivity debate is not simply whether AI replaces workers. It is whether the employer can lawfully manage the transition from the old job to the new one. A business that treats AI adoption as only an IT procurement may discover too late that it was also an employment restructuring, a monitoring programme and a data protection project. The safer approach is to recognise those legal consequences before the system begins making decisions about people.

Source note. This article is based on the International Labour Organization, Generative AI and Jobs: A Refined Global Index of Occupational Exposure (2025), Generative AI and Jobs: A 2025 Update, Algorithmic Management in the Workplace, and AI in Human Resource Management: The Limits of Empiricism (2025); Kenya’s Employment Act, especially section 40; Kenya’s Data Protection Act, especially section 35, and the Data Protection (General) Regulations, especially regulations 22 and 49; Uganda’s Employment Act as amended in 2026, especially sections 64 and 80; Tanzania’s Employment and Labour Relations Act, Cap. 366 R.E. 2023, especially the rules on unfair termination and retrenchment for operational requirements; Rwanda’s Law No. 66/2018 regulating labour, especially Article 22; and Rwanda’s Law No. 058/2021 relating to the Protection of Personal Data and Privacy, especially Articles 21 and 38.

Suggested citation: 

Ronald Serwanga, “AI at Work in East Africa: When Jobs Change by Law?” East Africa Legal Insight (5 September 2026).