More than 1,000 software engineers at Standard Bank are now using Amazon Bedrock and Q Developer, and Deputy Group CIO Khomotso Molabe told AWS Summit Johannesburg that early use cases had doubled the speed at which some teams deliver software. For AWS, Standard Bank is useful evidence for a much broader argument: Africa doesn’t need to spend decades retracing the technological development of richer economies before it can take advantage of AI.
The continent has skipped stages before. Jyoti Ball, General Manager for Sub-Saharan Africa at AWS, pointed to the spread of mobile phones without ubiquitous fixed-line networks and the growth of mobile money where conventional banking hadn’t reached much of the population. AI, she argued, could be another opportunity to jump ahead. She then moved to the other proposition technology companies increasingly bring to Africa: that AI can help solve problems that have resisted older technologies and institutions. Ball talked about language models handling customer service in African languages, computer vision identifying crop disease and developers building products more quickly. “AI will solve real problems on the continent, for the continent,” she said.
AWS had examples that made the pitch less theoretical. Shop2Shop is using an agentic Know Your Customer system in South Africa’s informal retail market, Jubilee Insurance has built an AI claims agent, and Aerobotics analyses drone, satellite and mobile imagery to monitor the health of 65 million trees. These are useful applications, but they’re a long way from establishing that AI can solve Africa’s problems.
Financial exclusion didn’t emerge because software wasn’t intelligent enough. Neither did poverty, unemployment or failing public institutions. Better technology can change what’s possible within those systems, sometimes considerably, without addressing why people lacked access, employment or effective services in the first place. AI companies nevertheless talk about Africa as though its problems are unusually well suited to technology. Less often discussed is whether the technology, the economic model behind it and the dependencies created by adopting it are particularly well suited to Africa.
That question becomes harder to avoid as the commercial case for AI moves beyond putting a clever assistant on somebody’s laptop. Companies are being told that agents can allow them to produce more software, process more work and automate tasks that previously required people. On a continent that needs to create jobs on an enormous scale, productivity has consequences that don’t fit neatly into a keynote slide.
Standard Bank itself shows why those consequences aren’t straightforward. Molabe said 80% of the bank’s migratable workloads are now running on AWS, its customer platforms have moved there and it has completed an internal AI Foundry through which engineering and solution teams can use the technology. The bank’s ability to deploy AI at this scale partly reflects years of work that came before it. When Tanuja Randery, AWS Vice President and Managing Director for EMEA, asked what had allowed Standard Bank to move beyond experiments, Molabe repeatedly returned to foundations and changes in how the bank works.
That complicates the leapfrogging story. A company founded in 2026 can avoid accumulating some of the legacy technology sitting inside an institution such as Standard Bank. An existing African bank can’t simply skip its way out of several decades of software. Standard Bank is moving quickly with AI partly because it spent a long time doing the less glamorous work required to make that possible.
FNB offered another view of the productivity argument. Chief Digital Officer Kevin Mitchell said the bank has about 3,000 experienced engineers and still has more technology demands than those engineers can meet. For a company in that position, giving developers better tools needn’t translate immediately into fewer developers because the backlog already exists.
Jonathan Allen, Executive in Residence at AWS, later gave an example with rather different arithmetic. Deriv expected a new charting platform supporting more than 170 indicators to require about 12 developers and between 12 and 24 months. AWS says one engineer using Q Developer built it in eight weeks. Deriv got its software faster and could put the rest of the team elsewhere. Another company budgeting its next project may notice that something once scoped for 12 people was completed by one.
Amazon has already acknowledged that AI productivity can eventually mean fewer jobs. In a message to employees, chief executive Andy Jassy said wider use of generative AI and agents would mean fewer people doing some existing jobs and more people doing others, but that Amazon expected its total corporate workforce to shrink over the next few years as it captured efficiency gains from AI.
Amazon subsequently announced reductions affecting about 14,000 corporate roles in October 2025 and another 16,000 in January 2026, although the company attributed those restructuring decisions more broadly to reducing layers and bureaucracy rather than saying those jobs had simply been replaced by AI.
What Amazon does doesn’t tell us what Standard Bank, FNB or another African employer will do, but Jassy’s comments remove the convenient assumption that displacement is only a theoretical concern raised by people who are sceptical of AI.
The IMF estimates that about 22% of jobs in Sub-Saharan Africa are likely to be affected by AI. Roughly 13% of total employment falls into work at significant risk of replacement, while almost 10% could benefit from AI complementing workers. The region’s relatively low exposure compared with wealthier economies may protect it from some immediate disruption, but the IMF notes that it could also mean Africa captures less of AI’s productivity and growth gains. By 2030, Sub-Saharan Africa is expected to account for roughly half of all new entrants into the global labour force, equivalent to as many as 15 million new jobs each year.
A project moving from 12 developers to one doesn’t tell us what happens to employment. The company might build more and keep everyone. It might redirect the other engineers towards work that was previously impossible to get to. It might also hire differently the next time around. The technology changes the calculation even when it doesn’t dictate the answer.
AWS has been thinking about another consequence of cheaper software for some time. Technical debt came up repeatedly at re:Invent 2025 in Las Vegas, where AWS Transform custom was pitched at old frameworks, outdated runtimes, API migrations and other work companies postpone because there’s always something more urgent to build. AWS says organisations allocate 20% to 30% of their software development resources to repeatable transformation work and claims Transform can cut execution time for many of those tasks by more than 80%.
AWS had already begun describing a broader version of this strategy at re:Invent 2025, where the launch of Kiro, AWS Security Agent and AWS DevOps Agent as frontier agents showed how quickly the company was moving beyond assistants towards systems expected to carry out larger pieces of work. Johannesburg was the continuation of that strategy, this time with African companies used to show what it looks like in practice.
In Johannesburg, Allen returned to technical debt, but the problem had changed. He said AWS Transform customers had eliminated more than 1.6 million hours of manual modernisation work, then argued that periodically cleaning up a codebase was no longer enough because AI-assisted development was increasing the rate at which software could be produced. Transform Continuous Modernization is supposed to work alongside development, finding outdated dependencies and other problems rather than waiting for a company to begin another large modernisation project.
Allen invoked Jevons’ paradox while talking about code generation: making something cheaper can lead people to use more of it rather than simply banking the saving. That is a plausible future for software because if code becomes much cheaper to produce, companies are unlikely to decide they already have enough. They can attempt products that were previously too expensive, add features that never made it through the backlog and build more internal systems.
The technical-debt pitch has therefore shifted in less than a year. At re:Invent 2025, AI was being used to deal with debt accumulated during years of conventional software development. In Johannesburg, AWS was describing AI that continuously deals with debt in an environment where AI-assisted development itself allows much more code to be produced. The old maintenance work hasn’t vanished because code became easier to generate. There may simply be much more software requiring it.
There’s a similar gap between having access to AI in Africa and having AI that has been properly adapted for African conditions. Language is the easiest place to see it. AfroBench was created to evaluate large language models across 64 Indigenous African languages and 15 tasks at a time when comprehensive evaluation of African languages remained limited.
South Africa presents a more complicated problem than adding isiZulu, isiXhosa or Afrikaans to a supported-language list. People switch languages halfway through conversations and sometimes halfway through sentences. Slang, borrowed vocabulary and cultural knowledge shape meaning. A customer-service agent can produce grammatically correct isiZulu and still misunderstand the person using it.
The problem isn’t limited to language. A model can know what a spaza shop is without knowing much about how one operates, how informal credit works between a shopkeeper and customers or why documentation expected from a formal retailer may simply not exist. African markets also differ enormously from one another. Credit histories, government processes, regulatory systems, public records and available datasets don’t become interchangeable because a product is marketed for “Africa”.
AWS itself spent a substantial portion of its keynote making a version of this argument about companies. Allen said “context is everything in an agentic world” because even a highly capable model doesn’t know how a particular organisation works. It doesn’t know which database is authoritative, whether a policy has been superseded or what somebody is permitted to see.
Amazon Bedrock Managed Knowledge Base is intended to connect agents to company information while maintaining document permissions, and its agentic retrieval system can evaluate information across different sources rather than simply returning whatever looks most similar to the question. AWS Context builds a knowledge graph from structured and unstructured organisational data and learns which sources tend to produce useful answers.
The reasoning is sensible: a generic model doesn’t understand Standard Bank simply because it understands banking. It needs Standard Bank’s accumulated knowledge, rules and exceptions. The same logic applies outside the company. An AI system doesn’t understand South Africa because it can write a paragraph about Johannesburg. Making it genuinely useful across South African languages, informal economies, financial arrangements and institutions requires local context that somebody has to collect, structure and maintain.
That work also raises a question about who supplies the layers around it. Ball said she wanted Africans not simply to consume AI but to “build AI and own AI”. AWS says it has invested $890 million on the continent since 2018, committed another $1.5 billion, brought 154 services to its Cape Town region and trained one million people in cloud and AI skills. Those investments give African companies access to computing capabilities that would be prohibitively expensive to reproduce independently.
Ownership becomes less clear as more of the AI stack comes from the same provider. A company may run its compute on AWS and access models through Bedrock. Agent Core can provide the environment in which its agents operate, Managed Knowledge Base can connect them to company information and AWS Context can organise it. Developers can use Q Developer, DevOps Agent can participate in delivering their software, Transform can maintain it and Continuum can work across security.
AWS also offers models from other AI companies through Bedrock. Allen argued in Johannesburg that organisations shouldn’t expect one model to be best at every job, which makes sense in a market where capabilities change quickly. But being able to change the model doesn’t necessarily remove the dependency if the compute, data connections, agent infrastructure, developer tools and security systems remain in the same cloud.
Cloud computing has always created some of this dependence. Migrating away from a provider is technically possible but can become extremely expensive once years of integrations, skills, architecture and internal processes have accumulated around it. AI creates more places for those attachments to form, particularly if agents become the way employees interact with company data rather than another workload quietly running in a data centre.
An African company can therefore become considerably more capable at building with AI while becoming more reliant on foreign technology infrastructure. It may own the application, the data and the customer relationship while finding that replacing the systems through which all three operate would be deeply disruptive. Few contemporary technology companies operate without depending on infrastructure owned by somebody else, but “Africans building AI” and Africa controlling more of its technological future aren’t necessarily the same thing.
Allen’s presentation suggested that AWS expects these relationships to deepen. He was dismissive of many of the enterprise assistants built during the first wave of generative AI, describing them as a “slightly faster search bar” because the person still had to move between systems and turn the answer into work. Amazon Quick is designed to take on more of that work, while Q Developer extends further into software development, DevOps Agent handles parts of what follows, Transform works on the existing codebase and Agent Core provides infrastructure for agents companies build themselves.
AWS is betting that African companies will want to move quickly enough that the trade-offs become secondary to the opportunity. Some probably will. What remains unresolved is whether the productivity gains create enough new work to offset what disappears, whether AI systems become meaningfully better at African languages and contexts, and how much technological control African businesses retain once more of their infrastructure, data access and software development depends on a handful of global platforms.
The Johannesburg summit didn’t answer any of that, but, it did make clear that AWS wants to be one of the companies (if not the company) on which those answers are increasingly reliant on..