Opinions expressed by Digital Journal contributors are their own.
Artificial intelligence has moved from experimentation into active enterprise investment. Boards are demanding it. Investors are funding it. Enterprises are budgeting for it. Yet many companies still struggle to show sustained, measurable return on investment from their AI initiatives. According to Johann Beukes, Chief AI Officer at Svitla Systems, the problem is not the technology. It’s maturity.
“We’re treating AI like traditional software,” Beukes says. “And it isn’t.”
Beukes has spent much of his career deploying AI systems inside enterprise environments, including data platforms, robotics applications, and computer vision systems. His background spans Fortune 500 consulting, venture-backed startups, and technical leadership roles that required turning experimental AI models into operational infrastructure. What he sees today is a structural misunderstanding.
AI isn’t a product, it’s a capability
Most executives still expect AI to behave like conventional software: scoped, delivered, and complete. But AI systems do not launch fully formed. They improve through exposure, feedback loops, retraining, and operational tuning. They evolve, and they require ongoing investment after deployment.
“The assumption is that once it works in a demo, it’s ready,” Beukes explains. “In reality, that’s when the work begins.”
This expectation gap quietly undermines ROI. Companies budget for development but often underestimate the ongoing investment required after deployment. They invest in proof-of-concepts, but not in scale architecture. They purchase licenses, but not internal capability. The result is stalled momentum.
The demo illusion
If there is one phase Beukes believes creates the most false confidence, it’s the proof-of-concept. In controlled environments, AI models can appear remarkably capable. But production environments introduce messy realities: data variability, edge cases, operational constraints, and nonlinear complexity growth.
A computer vision system trained to detect vehicle damage may perform well in a lab. In production, it must handle different lighting conditions, vehicle models, modifications, and unexpected anomalies. What appears to be incremental complexity quickly multiplies as real-world conditions introduce new edge cases.
Organizations that do not architect for scale from the outset often discover that their early success does not translate into enterprise viability. In many cases, the real bottleneck is not the model but the data. Organizations often pursue AI initiatives before their data pipelines and data quality are mature enough to support production systems.
Strategy before technology
Another pattern Beukes sees repeatedly is misaligned leadership priorities. AI initiatives frequently begin as innovation mandates, disconnected from revenue drivers or cost levers.
The companies that succeed, however, rarely frame AI as the centerpiece. Instead, they embed it within existing strengths. Logistics become smarter. Forecasting becomes more precise. Customer experience becomes more responsive.
“AI should strengthen the core business,” Beukes says. “Not compete with it.”
At Svitla Systems, his work increasingly focuses on helping enterprises connect AI capabilities directly to measurable outcomes, not just experimentation, but operational impact.
The governance gap
Perhaps the most underestimated risk in enterprise AI is governance. Unlike traditional software, AI systems influence decisions and, in some cases, operate with increasing levels of autonomy. Without structured oversight, monitoring, and accountability, organizations can unknowingly introduce systemic risk.
As agent-based systems gain autonomy, governance will become less about compliance documentation and more about operational design. Audit trails, model transparency, lifecycle management, and defined accountability structures will determine which organizations scale responsibly.
The shift to intelligent automation
Over the next three years, Beukes expects a shift from basic automation toward intelligent, agent-driven workflows. But the deeper transformation will not be technical, it will be organizational.
“Technology adoption will outpace human readiness,” he predicts.
Roles will evolve. Technical leaders will move from writing code to orchestrating AI systems. Workflow design, lifecycle oversight, and cross-agent coordination will become core skills. Beukes describes the emerging archetype as the “agent manager”, professionals who supervise intelligent systems rather than manually execute every task. At the enterprise level, this shift may allow teams to produce significantly more output with the same or smaller groups. AI, when deployed strategically, acts as a multiplier.
“The smartest companies aren’t replacing people,” he says. “They’re amplifying them.”
Training before tools
Despite rapid enterprise investment in AI platforms, one of the most overlooked variables remains education.
Many professionals still use only a fraction of what modern AI tools can enable. Adoption is not merely a procurement decision, it is a behavioral shift. Without training and structured integration into workflows, even powerful systems remain underutilized. The next competitive divide will not be defined by which company bought AI first. It will be defined by which organization built the discipline, governance, and expertise to use it effectively.
AI is no longer a novelty advantage. It is an operational capability, and like any capability, it rewards companies willing to treat it with long-term rigor.
