This Kenyan startup wants to rebuild enterprise software around AI agents

In most companies, software is still built around the logic that work is broken into steps, passed through systems, and tracked by people. Kenyan AI startup Lua thinks that logic is becoming obsolete.

The Nairobi-based startup has raised $5.8 million (about KES 748m) in seed funding to expand a platform for building AI “agents,” software systems designed not to assist employees, but to carry out entire business processes end-to-end. Norrsken22 led the round, with participation from Y Combinator, Flourish Ventures, P1 Ventures, and Enza Capital.

Lua’s pitch reveals a shift in enterprise AI. After nearly five years of chatbots, copilots, and productivity tools embedded in existing systems, some startups are now betting on software that performs work rather than supporting it.

Founded in 2023, the startup provides a platform that allows organisations to create and deploy autonomous agents that can handle multi-step workflows such as customer onboarding, loan processing, and claims management, while operating through existing business channels like Slack, WhatsApp, and email.

“We’re in the race to shape how human-agent collaboration gets defined globally,” says co-founder Lorcan O’Cathain. “Organisations will be blends of humans and AI agents collaborating.”

Enterprise software has long been designed around decomposition.

For example, a customer application is split into verification, risk assessment, approval, and onboarding. Each step sits in a different system or team. Even where automation exists, it tends to optimise within those boundaries rather than remove them.

Lua argues that AI changes the constraint that made that structure necessary.

Its platform allows companies to build agents that can take a request—for example, a loan application—and carry it through multiple stages without being handed off from system to system. The agent collects data, checks conditions, applies rules, and escalates only when uncertainty is high.

The company’s systems operate through existing business channels, including WhatsApp, email, and voice, rather than requiring firms to adopt new interfaces.

In early deployments in Kenya, most use cases sit in financial services, where manual processing delays are a persistent operational constraint. Some Kenyan banks take 3 to 5 days on average to process unsecured retail loans, with manual KYC checks and document verification driving most delays

“One of the most valuable skills someone will have will be the ability to manage agents and help improve them,” O’Cathain says. In his view, the boundary between technical and non-technical teams is blurring as agents take on execution-heavy work across functions.

Already, early users are experimenting with small teams supervising large numbers of agents—in some cases, 1–3 people coordinating “10+ agents”—or structuring agents as the core of the product itself.

“It’s exciting to see how many small teams across the continent are looking to scale their business with agents,” he says.

For now, most enterprise AI tools remain assistive: summarising documents, drafting responses, or generating code suggestions, with platforms from Microsoft, Google Cloud, and Amazon Web Services leading that layer, while automation firms like UiPath extend existing workflows. Lua is targeting a more ambitious threshold: systems that can reliably complete structured business processes with minimal human input, a model that could appeal to Kenyan companies still dealing with manual back-office operations and fragmented systems.

“A year or two ago, AI could summarise documents and generate text, but it couldn’t reliably execute multi-step workflows with real business logic,” O’Cathain says. “That changed fast.”

He argues that this shift is already reshaping corporate adoption patterns. A recent study by PwC suggests 79% of US companies are now actively exploring AI integration, though most remain in early deployment phases.

For Lua, the opportunity lies in moving those experiments from augmentation to execution.

The company describes its platform as turning “an LLM into a fully functioning employee inside your org chart,” a framing that reflects how quickly AI is being mapped onto organisational structures rather than just tools.

Despite the language of autonomy, Lua’s system is not designed to operate without people. It is structured around handoffs, where agents escalate cases that fall outside confidence thresholds or regulatory boundaries. Human staff remains responsible for oversight, exception handling, and final approval in sensitive workflows.

That design reflects both technical constraints and sector requirements. In banking and insurance, full automation is still difficult to reconcile with compliance obligations and risk management frameworks.

But even partial delegation changes how work is organised. Instead of employees processing individual cases, early users describe shifting towards supervising multiple automated processes at once, monitoring exceptions, correcting errors, and refining system behaviour. In effect, the unit of work moves from execution to oversight.

Lua’s approach has been shaped by its environment as much as its technology.

In Kenya, enterprise software adoption is driven by immediate operational pain rather than long-term transformation narratives. That has pushed the company towards use cases where outcomes are measurable, such as faster onboarding, shorter approval cycles, and reduced manual queues.

O’Cathain says it has also influenced product design. Rather than replacing existing systems, Lua integrates with tools businesses already use, particularly messaging platforms that dominate day-to-day operations, such as WhatsApp and Facebook.

The company says it also gives clients direct control over their agents, a choice it believes is important in markets where trust in software vendors is still evolving.