The distinction matters. Adoption stories are about efficiency gains and cost savings. Building stories are about what gets created, who owns it and where the value lands. Venture studios with small, opinionated teams that conceive and ship AI products as their core business represent a different bet on SA's position in the global AI stack. We've tracked SA's AI adoption story in depth. What follows is the other side: who's actually building.

These three studios aren't accelerators or funds. They don't take pitches and write cheques. They build. Here's who they are.

Adii Pienaar is a four-time founder. He co-founded WooCommerce, the WordPress e-commerce plugin that powers millions of stores globally, built and sold Conversio to Campaign Monitor for around R124 million, and co-founded Cogsy, a demand planning tool that was later acquired by Mayple. (We profiled Adii in our original piece on the Stellenbosch founders who've generated serious exits.) Between Cogsy and Ubundi, he spent eighteen months at Automattic as General Manager of its "Other Bets" division, overseeing a portfolio that included Tumblr, Pocket Casts, Day One and Gravatar, an unusual vantage point for a South African founder, sitting inside one of the world's most successful open-source companies while watching AI reshape the product landscape around him.

He left Automattic in June 2025 and launched Ubundi shortly after. The Cape Town-based AI product studio is building human-centred AI products grounded in a specific philosophical position: that the financial, intellectual and cultural value created by AI should flow first to South Africa, and that AI should amplify human context rather than replace it. The name is a deliberate invocation of Ubuntu, the Nguni philosophy that "I am because we are."

The lead product is TooToo, which Adii has described as a tool for building and maintaining an explicit representation of who you are, your values, beliefs, preferences and boundaries. The idea isn't a personality quiz or a one-time snapshot. It's an ongoing personal context layer: the kind of grounding that lets any AI tool know who it's talking to before it starts, regardless of which model is under the hood. We covered TooToo in more depth when it launched earlier this year. Ubundi has since spun out its second product, Kwanda, as a separate venture (kwanda.ai), focused on AI development tooling. The studio is also raising Fund 1 to back South African founders building in AI globally, moving from product builder to early-stage backer at the same time.

The question is whether a studio with this much surface area can execute across all of it simultaneously. Products, a spin-out, and a fund are a lot to hold together in year one. Adii's track record suggests he knows how to build things people actually use. Whether Ubundi's ambition matches its bandwidth is the thing to watch.

Where Ubundi is building end-user products, XGMI is building the infrastructure. The Johannesburg-based studio, the name stands for {x} General Machine Intelligence, is developing Agent Core, an AI agent platform designed as the infrastructure layer that other AI applications can run on top of.

The founding team is small, led by Tyler Reed, and the studio has articulated a three-phase roadmap. Phase one is AI co-pilots: tools that work alongside humans without adding complexity. Phase two moves to autonomous agents that handle tasks independently. Phase three is the ambitious one, autonomous solutions scaled across industries, interconnected into something approaching genuine machine intelligence at scale.

The roadmap isn't unusual for an AI infrastructure play. What's more distinctive is the design philosophy underneath it: XGMI has been explicit that simplicity and privacy are first-order constraints, not afterthoughts. In a market where most agent frameworks are built by engineers for engineers, "simple enough that humans can actually use this" is a real product bet.

The question XGMI hasn't fully answered publicly is what Agent Core does in concrete practice, which specific tasks it handles, what integration points it exposes, and whether there are clients or case studies that demonstrate the platform working in the wild. For an infrastructure play, that evidence is load-bearing. Platforms win by becoming the default; you become the default by having people build on you. Watch for whether XGMI starts naming what's been built on top.

Melio AI is the most immediately commercial of the three. The Johannesburg-based MLOps consulting and products firm launched Highwind in February 2025, positioning it as Africa's first AI marketplace, a platform where businesses can access ready-to-deploy AI tools without building or maintaining them in-house. Entry pricing starts at R900 per month.

The tools on the marketplace are built for the SA business context in ways that generic global platforms aren't. Document verification covers 18 document types, including CIPC filings and bank statements. The ID validation tool cross-checks against credit bureau databases. There's a handwritten form data extractor, a knowledge base builder for RAG applications, and a customer support agent that chains multiple Highwind tools together into a single workflow. The marketplace currently spans seven industry verticals: financial, technology, logistics, healthcare, legal, retail and manufacturing.

The mechanic that makes this a studio model, not just a marketplace, is the request function. If a business needs an AI tool that doesn't exist on Highwind yet, they can request it. Melio builds it, and the new tool goes onto the marketplace for others to use. It's demand-driven product development: the market signals what to build, the studio builds it, and the result becomes a scalable product rather than a one-off client deliverable.

This is a smart structural solution to a real problem. Most AI consulting businesses face a ceiling: revenue is tied to hours, scope creep is constant and IP stays with clients. Melio's model is different. Client demand funds product development. Products accumulate on the marketplace. Over time, recurring marketplace revenue offsets the project work. It's not a new model, it's the classic services-to-products transition, but applied specifically to SA's small business market, where the need for affordable, locally relevant AI is real and underprovided.

The open question is speed: how quickly the marketplace reaches the critical mass of tools and buyers that makes the model self-reinforcing. The sectors are right. The pricing is accessible. The request mechanic is clever. Traction numbers would tell the rest of the story.

The AI venture studio model is emerging beyond SA, too. Amunai, operating across Africa, the Middle East, and Europe, claims to be the world's first AI-native venture studio, with $25 million in funding and a proprietary agentic AI framework spanning more than twenty verticals. Revise takes a different approach: it co-designs AI companies focused on African markets, emphasising product-market fit and tackling themes such as climate, mobility, and future finance.

Both are worth watching. The three SA studios profiled here are doing something more specific: building for and from the South African context, with SA founders, SA capital, and an explicit stake in where the value ends up.

Are you building an AI venture studio in South Africa or Africa? Your name should be up here, then. Get in touch to fix that quick.

This news first appeared in our 15 Apr ‘26 newsletter on how SA makes money off of property.

Read our original feature on the Stellenbosch founders who built and exited companies worth hundreds of millions. We covered TooToo and the AI context problem in depth when Ubundi launched earlier this year. See the broader picture of AI adoption in South Africa in 2026.

Get more SA tech and business news and subscribe to The Open Letter.