Enterprise AI has become remarkably easy to buy and stubbornly difficult to make useful. A company can give thousands of employees access to ChatGPT, Copilot or another AI assistant relatively quickly. Connecting those systems to years of institutional knowledge, existing access controls and the messy way work actually moves through an organisation is where the difficult part begins.
That gap is what South African-founded startup Verascient is trying to turn into a business. The company has raised US$1.2 million, around R19.5 million, in its first funding round to develop what it describes as an operating system for companies that want AI to become part of how they work rather than another application employees occasionally open.
The oversubscribed round was backed by Founder Collective, whose portfolio has included Uber, Airtable and Whoop, as well as Andrena Ventures, Cambridge Enterprise and Summit Ventures. South African entrepreneur Alan Knott-Craig and other angel investors also participated. Verascient says the money will be used to expand its engineering team and develop its technology for more enterprise deployments.
The funding number is relatively small beside the billions being poured into AI infrastructure globally, but Verascient is aiming at a problem that has become increasingly visible as companies move beyond the first few years of generative AI experimentation: having capable models doesn’t automatically give those models an understanding of a business.
McKinsey’s 2025 global AI survey found that 88% of respondents said their organisations were using AI in at least one business function, yet nearly two-thirds hadn’t started scaling it across the enterprise. Just 39% reported any impact on earnings before interest and tax from AI. The gap between using AI somewhere in a company and changing how the company operates has become large enough to support an entire new category of enterprise software.
Giving AI some institutional memory
Verascient’s approach begins with something it calls a temporal knowledge graph. The idea is to collect knowledge that would otherwise sit across documents, spreadsheets, internal systems and individual employees, while retaining information about where it came from, who should be allowed to access it and how it has changed over time. AI agents can then operate against that organisational context instead of beginning every task with whatever happens to be contained in a prompt.
Co-founder and CTO Emile Ferreira compares the architecture with an operating system. In his analogy, the knowledge graph functions like a file system, while the company’s agent-to-agent protocol allows different AI processes to communicate and background agents can perform work without waiting for an employee to prompt them. Verascient also says its platform has more than 1,000 integrations with other applications and services.
There is a sensible idea underneath the operating-system language. The usefulness of an AI agent inside an insurer, bank or logistics company depends heavily on whether it can retrieve the correct information at the right moment without wandering into data it shouldn’t be able to see. Once agents can also take actions rather than merely generate text, provenance and permissions become operational questions rather than abstract governance concerns.
South African businesses appear keen to get there. The SAS Data and AI Impact Report found that 91.2% of its South African respondents were using generative AI and 67.6% reported using agentic AI, ahead of the global figures in the survey. Yet only 8% described their AI maturity as transformative, while just 6% said their data infrastructure had reached the report’s highest “optimised” level.
Those numbers are survey responses rather than a census of South African business, but the mismatch is useful. Companies can adopt AI faster than they can clean up the systems, knowledge and processes the technology eventually needs to interact with.
The AI company hiring humans to make AI work
Verascient’s other bet is that software alone won’t close that gap. Its deployments are paired with AI engineers who work directly with customers to identify processes that can be redesigned and then build the systems around them. The company is initially concentrating on financial services, insurance and logistics, where institutional knowledge can be spread across years of records and internal systems.
That approach is becoming increasingly familiar in enterprise AI. AWS recently committed $1 billion to a Forward Deployed Engineering organisation that will place engineers inside customer teams to help build AI systems around their data, governance and existing business processes. The scale is completely different, but the underlying diagnosis is strikingly similar: increasingly capable AI hasn’t removed the need for people who understand how to make it work inside an organisation.
The comparison is particularly interesting because AWS is doing this while simultaneously selling increasingly autonomous AI systems. Its decision to spend $1 billion putting skilled people closer to customers suggests that the difficult part of enterprise AI has shifted. Model capability still matters, but organisational knowledge, workflow design, governance and implementation are becoming harder constraints.
For Verascient, that makes the human component commercially useful now, but it creates a question the company will eventually have to answer. A high-touch model can solve messy customer problems that packaged software can’t, particularly while enterprise AI practices are still immature. As the company grows, the test will be whether the engineering work makes its platform progressively easier to deploy or whether every new customer continues to require substantial bespoke implementation.
South African talent, global customers
Verascient is using some of its funding to recruit what CEO and co-founder Keagan Stokoe describes as South Africa’s “top 1%” of AI talent. The company wants a relatively small engineering team working on deployments for customers beyond South Africa, effectively treating local technical talent as an export rather than assuming the company itself needs to move its centre of gravity elsewhere.
Both founders arrive with experience that makes the pitch less theoretical. Ferreira was an early developer at Replit before completing an MPhil in Advanced Computer Science at the University of Cambridge. He later co-authored research with the United Nations International Computing Centre examining how organisations should measure the full cost and value of AI deployments.
That research, conducted with Cambridge’s Frugal AI Hub and UNICC, argues that organisations still struggle to measure AI beyond individual projects, proposing a framework that accounts for total cost of ownership, return on investment and wider organisational impact. The connection to Verascient’s thesis is fairly direct: enterprises are moving from experimenting with individual AI tools towards having to understand AI as part of the organisation itself.
Stokoe previously worked on the founding team at South African fibre company Fibertime before starting AI consultancy Purple Dorm, which worked with organisations in South Africa and the UK.
He also presents Verascient’s hiring plans as a counterpoint to fears that AI adoption will inevitably mean fewer opportunities for people. The company’s own hiring supports that argument only up to a point. Demand for engineers who can build and deploy these systems can grow at the same time that AI reduces the amount of human work required elsewhere in an organisation. One startup recruiting highly skilled engineers doesn’t settle the broader employment question.
What it does show is where some of the value created by enterprise AI may accrue. If companies increasingly need people who can connect models to data, permissions, processes and business objectives, countries that produce those skills have something more valuable to export than inexpensive technical labour.
For now, Verascient hasn’t disclosed customer names, revenue figures or enough deployment data to judge how much measurable business value its systems are producing. The R19.5 million round shows that investors are willing to back its diagnosis of the enterprise AI problem, but the more consequential proof will come from what happens inside the companies using it.
Enterprise AI may be becoming less about access to intelligence and more about access to context. Models from OpenAI, Anthropic, Google and others will keep improving and, in many cases, become interchangeable for particular tasks. A company’s accumulated knowledge, permissions, processes and peculiar way of getting work done aren’t interchangeable at all. Verascient is betting that building a system capable of understanding that layer is where the harder and more defensible business lies.