According to a BusinessTech article, Codehesion, described as South Africa’s top software development company in the 2026 MyBroadband Awards, says its use of artificial intelligence (AI) has made low-code development virtually obsolete for businesses that need custom software.
Low-code platforms have traditionally allowed organisations to build applications with minimal coding, relying on visual drag-and-drop tools. The promise was faster delivery and reduced need for large development teams. In the Codehesion statement, the company argues that AI tools now automate many of the coding steps that low-code platforms used to simplify, delivering faster and cheaper outcomes.
What AI brings to the development process
AI-augmented development typically involves large language models that can generate code snippets, suggest fixes, and even write whole modules based on natural-language prompts. Tools such as GitHub Copilot, Amazon CodeWhisperer and local open-source models are being integrated into development pipelines to speed up routine tasks. For a South African firm, the benefit is twofold: reduced labour costs and a shorter time-to-market, both of which are critical for small and medium-size enterprises (SMEs) that operate on tight budgets.
Codehesion claims that by embedding these AI capabilities into its workflow, it can shrink project timelines and lower overall costs compared with the recurring subscription fees that low-code vendors charge. The company also notes that the software it delivers is fully owned by the client, with an optional maintenance service.
For an SME owner, the decision now hinges on whether a subscription-based low-code platform or a custom AI-enhanced build offers better value. Low-code licences can run into tens of thousands of rand per year, especially for enterprise-grade features. A custom project, even with AI assistance, typically involves a one-off development fee plus any agreed-upon support costs. The claim that the latter is always cheaper is a point that would need verification against actual pricing data from both models.
The broader South African tech sector is seeing a rise in AI adoption. Government programmes such as the Digital Skills Programme and private investment in AI startups have increased the pool of talent capable of building AI-driven solutions. At the same time, the low-code market remains active, with platforms like Mendix and OutSystems maintaining a presence in larger corporations that value rapid prototyping and internal citizen-developer initiatives.
Industry observers note that the choice between low-code and custom AI-augmented development is not purely cost-driven. Low-code tools often provide built-in governance, security and compliance features that can be attractive for regulated industries. Custom builds, even when accelerated by AI, still require rigorous testing and may need additional security reviews.
Codehesion’s claim that there is “no metric where low-code makes sense” is a strong marketing line. Independent analysts would look for comparative case studies that measure development time, total cost of ownership and post-deployment performance across both approaches. Until such data is publicly available, the statement remains a company claim.
What is clear, however, is that AI is reshaping how software is built in South Africa. Companies that can combine AI tools with experienced development teams are likely to offer more competitive pricing and faster delivery than those relying solely on low-code platforms. For business owners, the practical step is to request detailed proposals that break down AI-related efficiencies, compare them with low-code subscription costs, and assess the long-term support model.
The part of software development that AI has not compressed
Generated code still has to be read, and that is where the economics of this argument are actually settled. A model can produce a working function far faster than a person can type one, but somebody with the relevant expertise still has to confirm that it does what was intended, that it handles the cases nobody thought to prompt for, and that it has not quietly introduced a security weakness. Review does not scale the way generation does, which is why teams adopting these tools often find that the bottleneck moves rather than disappears.
The comparison with low code also depends on which cost is being counted. Licence fees are visible and they recur, so they are easy to put in a proposal. The costs that decide the total over several years are usually less visible: what it costs to change the system when the business changes, whether the work can be handed to a different supplier, and what happens to the application if the platform vendor raises prices, changes direction or is acquired. Custom code avoids the platform dependency and replaces it with a maintenance obligation, which is a different risk rather than an absent one.
Governance is the third variable. Regulated sectors have to demonstrate who changed what and why, and an audit trail is easier to produce when the tooling was built to produce one. A custom build can meet the same standard, but the effort is explicit rather than included in the licence.
For a buyer, the useful move is to make the comparison concrete rather than philosophical. Ask what a defined change will cost a year or two after delivery. Ask what the handover looks like if the relationship ends. Ask who reviews AI generated code before it ships, and what happens when the review finds something. Vendors on both sides of this argument can answer those questions, and the answers, rather than the positioning, are where the real difference shows up.
