“South African businesses are approaching a new phase of artificial intelligence in which AI agents can carry out complex assignments, coordinate activities across digital applications and work for extended periods with less step-by-step human direction. Google Cloud executive Maureen Costello says organisations must redesign their workflows, prepare employees for changing responsibilities and establish effective governance rather than simply adding AI tools to existing operations.”

South African businesses face a new workplace transformation as AI agents take on complex tasks

Artificial intelligence is moving into a new stage of workplace adoption, with autonomous AI agents increasingly capable of performing complex assignments rather than simply answering questions, generating text or responding to individual prompts. For South African businesses, this development presents opportunities to improve productivity and service delivery, while raising important questions about employment, skills, accountability and the future organisation of work.

The shift was highlighted by Maureen Costello, vice-president for the United Kingdom, Ireland and Sub-Saharan Africa at Google Cloud, during the Singularity Summit South Africa 2026. In a presentation about the risks of assuming that technology alone can transform an organisation, Costello argued that businesses must reconsider how work is structured as AI systems become capable of handling more complicated responsibilities. ITWeb reported on her remarks on 26 October 2026.

The central message is that the next phase of AI adoption will require more than purchasing software. Organisations will need to rethink processes, establish appropriate oversight and ensure that employees understand how to work alongside increasingly capable digital systems.

From digital tools to AI assistants and autonomous agents

The evolution of workplace technology can be understood through three broad stages.

The first stage involved conventional software. Employees used applications to perform defined tasks, entering information, clicking through menus and following established procedures. Although this software improved efficiency, people generally remained responsible for coordinating the different steps required to complete a job.

The second stage introduced generative AI assistants. These systems can draft correspondence, summarise documents, analyse information, develop computer code and answer questions. However, employees typically need to provide instructions, review the output and direct the next step.

The emerging third stage involves AI agents that can pursue a defined objective across multiple steps. Depending on their design, permissions and connected applications, they may gather information, prepare reports, coordinate activities and carry out routine actions without requiring a new instruction for every stage.

Costello described this transition as a fundamental change in the relationship between workers and technology. Instead of directing every individual operation, employees could increasingly delegate assignments and supervise their completion.

For example, a business might use an AI agent to assemble a weekly sales report. The system could retrieve authorised sales information, compare performance across branches, identify unusual changes and prepare a summary for a manager. A human would still need to check the findings, confirm their accuracy and decide what action to take.

The important distinction is that an agent may coordinate several related tasks towards one objective. Nevertheless, the extent of its independence depends on the system, the quality of available data and the permissions granted by the organisation.

Why South African companies should pay attention

South African businesses operate in an environment where productivity, operating costs, customer service and access to specialised skills are significant considerations. Properly implemented AI could help organisations reduce repetitive administrative work and allow employees to concentrate on activities requiring judgement, communication and problem-solving.

For smaller enterprises, the technology could offer practical benefits without requiring large teams for every business function. An appropriately configured system might assist with customer enquiries, prepare quotations for review, organise internal documents or identify patterns in business performance.

Larger organisations could explore AI agents for coordinating information across departments, supporting customer service operations and assisting employees with routine reporting. In sectors such as retail, financial services, logistics and telecommunications, these applications could help teams process information more quickly.

However, these benefits are not automatic. A poorly designed AI workflow can reproduce existing inefficiencies, introduce errors more rapidly or create additional work when employees must correct inaccurate outputs.

The strategic challenge is therefore to identify the business problem first and select technology that addresses it. Companies should establish measurable objectives, such as reducing processing delays, improving response times or lowering the frequency of administrative errors, before judging whether an AI system delivers value.

Why buying AI is not the same as transformation

One of the most important implications of Costello’s remarks is the need to reconsider existing business processes. Installing an AI assistant in an outdated workflow may speed up individual tasks without addressing unnecessary approvals, duplicated data entry or poor communication between departments.

A more meaningful transformation begins by examining how work is performed from start to finish. Managers can identify activities that are repetitive, determine which decisions require human judgement and establish where automation could safely reduce delays.

For instance, a company handling customer complaints might currently require an employee to collect information from several systems, categorise the complaint, forward it to the correct department and prepare a response. An AI agent could potentially assist with gathering authorised information, classifying the issue and preparing a draft response.

Yet decisions involving compensation, legal obligations, sensitive customer circumstances or significant financial consequences may require explicit human approval. The goal should be to remove unnecessary effort without weakening accountability.

This distinction matters because an organisation’s existing procedures often reflect historical decisions rather than the most effective way to achieve an outcome. AI adoption provides an opportunity to reassess those procedures instead of merely automating them.

The implications for jobs and employee skills

The expansion of AI agents is likely to change the composition of some jobs. Tasks involving routine document preparation, information retrieval, basic analysis and administrative coordination may increasingly be supported or performed by automated systems.

That does not mean every affected position will disappear. The consequences will depend on the occupation, the organisation’s technology, implementation costs and the degree to which tasks require human interaction or professional judgement.

Some employees may spend less time producing routine reports and more time interpreting findings, managing relationships or making decisions. Other roles could require substantial redesign, particularly where a large proportion of daily work involves predictable digital tasks.

South Africa’s employment challenges make this transition particularly important. If entry-level work changes substantially, employers and educators will need to consider how young people can gain practical experience and develop the capabilities needed for future jobs.

Training should include digital literacy, critical thinking, data protection, communication and the ability to evaluate AI-generated information. Employees also need to understand that automated output can be incomplete, biased or incorrect.

Businesses that communicate clearly about the purpose of AI adoption may be better positioned to build trust. Workers should understand which tasks are changing, how performance will be assessed and what opportunities exist to learn new skills.

Human oversight remains essential

Autonomous systems introduce risks that organisations must address before allowing them to perform important tasks. An AI agent could misinterpret an instruction, rely on inaccurate information, make an inappropriate recommendation or take an action outside its intended purpose if its permissions and safeguards are inadequate.

Businesses should therefore establish clear boundaries for automated decision-making. These may include restricting access to confidential information, requiring approval for high-impact actions, recording system activity and providing a process for investigating errors.

Data quality is equally important. If an AI system receives outdated customer records or incomplete financial information, it may produce convincing but misleading results. Organisations must ensure that data is relevant, properly managed and accessible only to authorised users.

Cybersecurity also requires attention. Connecting AI agents to email, customer databases, financial platforms and other business applications can increase the consequences of compromised accounts or poorly controlled access.

Effective governance should define who owns the system, who may authorise its actions and who is responsible when something goes wrong. Human oversight should be meaningful rather than a routine approval exercise in which employees accept recommendations without checking them.

Africa’s opportunity to build AI-enabled services

Costello also highlighted the opportunity for African organisations to develop AI-enabled operations and apply the technology to regional challenges. Potential applications include health research, support for medical screening, agricultural information and educational assistance in African languages.

These areas are relevant because services designed around local conditions may address needs that generic international systems do not adequately meet.

In agriculture, for example, AI-supported tools could help present weather information or farming guidance in accessible formats. In education, carefully designed language tools could supplement learning materials and provide additional practice opportunities. In healthcare, AI may assist qualified professionals with selected research and screening tasks, provided appropriate validation and clinical oversight are in place.

Realising these opportunities requires investment in connectivity, reliable electricity, suitable computing infrastructure, relevant data and local technical expertise. It also requires systems that reflect local languages, cultural contexts and practical operating conditions.

African organisations should not simply adopt imported technology without evaluating whether it meets their needs. Developing local expertise and participating in the design and governance of AI systems can help ensure that economic value and technological capabilities are built within the region.

A practical path forward for South African businesses

Companies considering autonomous AI should begin with limited, measurable projects. Suitable starting points include internal document search, routine report preparation and other tasks where errors can be detected before they cause significant harm.

Each project should have a clear purpose, defined access permissions and a named person responsible for reviewing its performance. Managers should compare results with the existing process, paying attention to accuracy, cost, processing time and employee experience.

Organisations should also involve workers in the redesign of their responsibilities. Employees often understand the practical difficulties of existing workflows and can help identify where automation will be useful or where it could create new risks.

As systems become more capable, companies should regularly review their permissions, test failure scenarios and update training. They should also provide customers with appropriate transparency when AI plays a material role in an interaction or decision.

Ultimately, the success of AI agents will not be measured solely by how much work they can perform independently. It will depend on whether they improve outcomes while preserving security, trust and human accountability.

Conclusion

The move towards AI agents marks a significant potential shift in how South African organisations structure work. Businesses may gain opportunities to automate complex routines, improve productivity and develop new services, but those gains will require deliberate changes to workflows, employee training and governance.

The message emerging from the Singularity Summit South Africa 2026 is clear: organisations should not treat AI as another software purchase. They should determine where it can create genuine value, redesign processes around clear objectives and ensure people retain appropriate control over important decisions.

For South Africa, the opportunity is to use this technological transition to strengthen local capabilities, improve services and prepare workers for changing demands. The challenge is to achieve those benefits without allowing speed and automation to outrun reliability, fairness and accountability.