Norrsken22 leads $5.8M seed for Lua to scale the first AI agent OS in Africa

Lua, a startup building software for managing AI agents, has raised $5.8 million in seed funding led by Norrsken22. The round includes participation from Flourish Ventures, 20VC, P1 Ventures, Phosphor Capital, and Y Combinator, alongside angel investors like Henri Stern and Kaz Nejatian.

The company launched its platform in October 2025 and says revenue has been growing at nearly 30% week-on-week since then. In February 2026, more agents were built on Lua than in all previous months combined, a signal of how quickly companies are moving from testing AI tools to deploying them inside workflows.

Early customers include African startups such as Turaco, Tushop, Umba, and Numida, which are using Lua to automate internal processes and coordinate tasks between human teams and AI systems.

The shift Lua is betting on is already visible. Companies are no longer struggling to access AI models. The problem now is making them useful inside actual work. Lua is positioning itself as that layer. Instead of plugging into separate tools or paying per API call, companies can deploy AI agents, assign them tasks, and run them alongside human teams from one system.

The platform handles infrastructure, model orchestration, integrations, and monitoring, leaving teams to focus on business logic.

“The companies that will win over the next few years are the ones that build their agent workforce with the same intentionality they bring to their human workforce,” said CEO Lorcan O’Cathain. “Most businesses are either blocked by technical complexity or locked into rigid tools that don’t reflect how their teams actually work. Most agent platforms compound this with black box tooling and per-outcome pricing: the more your agents succeed, the more you pay, with no pathway to improving your agent economics. Lua is built on the opposite principle: teams own their agents, own their outcomes, and build compounding efficiency over time.”

The product is designed for both technical and non-technical teams. Developers can also deploy agents through a command-line interface, while non-technical staff can use a visual interface to build and manage workflows. Both operate on the same system, allowing teams to collaborate around shared processes.

This dual approach reflects the founders’ background. O’Cathain previously led Zephyr’s Africa business, while Stefan Kruger was VP of Engineering at Paystack before its acquisition by Stripe. Both worked together scaling a fintech company in East Africa, where operational complexity often forces companies to build internal tools earlier than their global peers.

This influence shows in where Lua is gaining traction. Instead of targeting experimental AI use cases, the platform is being used in fintech, lending, and retail operations where tasks are repeatable and measurable. In these environments, small efficiency gains can directly affect revenue or cost.

Investors are leaning into that positioning. Norrsken22, which has backed companies like TymeBank and Sabi, is betting that AI infrastructure built for emerging markets can scale globally.

“We are thrilled to support Lua. The founders fundamentally understand how agent and human workforces need to collaborate to get work done,” said Lexi Novitske, General Partner at Norrsken22. “Additionally, they are a global company that has deployed in Africa, Asia, the U.S. and Europe with deep experience, a volume of data, and a pricing intuition that’s difficult to replicate. We’re excited to help them build out this operating system for human and agent workforces.”

The AI startup plans to use the funding to expand its developer community and grow its Lua Implementation Network, a group of partners deploying the platform across different markets.

The broader bet is straightforward. As companies move past experimenting with AI, the demand is shifting toward systems that can manage how that intelligence is used. Tools that organise agents, track performance, and control costs may end up becoming the layer where most of the value sits.

Last updated: April 17, 2026