The Evolution of AI-Driven Development
The term ‘vibe coding’—coined by OpenAI co-founder Andrej Karpathy in February 2025—is rapidly losing its status as a revolutionary development paradigm. While it initially promised to democratize software creation by allowing users to generate applications through natural language prompts, industry experts and Karpathy himself now view it as a transient ‘bridge phase’ toward more sophisticated methodologies.
Reports from Forbes and the Boston Consulting Group (BCG) suggest that the industry is shifting toward ‘agentic engineering.’ Unlike vibe coding, which relies on simple prompting and rapid scaffolding, agentic engineering focuses on directed autonomy, where AI agents manage complex workflows with minimal human intervention.
Risks in Finance and Compliance
The transition is not merely technical but deeply institutional. BCG analysts have warned that while AI coding agents can accelerate tasks like anomaly detection and financial forecasting, they introduce significant ‘shadow code’ risks. Without robust governance, organizations risk creating undocumented, unmanaged scripts that bypass official oversight, leading to potential auditability and compliance failures.
Legal experts, including those from Brownstein Hyatt Farber Schreck, have raised concerns about the enforceability of trade secrets within AI-generated systems. As organizations integrate these tools at scale, the lack of standardized governance frameworks remains a critical vulnerability.
The ‘Karpathy Factor’ and Future Outlook
The shift in nomenclature reflects a deeper change in strategy. Karpathy’s move to Anthropic to lead pretraining research signifies that the bottleneck in AI development has shifted from simple code generation to model capability. The focus is no longer on how quickly a user can ‘vibe’ a prototype, but on how effectively an agent can execute a task from start to finish without human correction. For businesses, this means the future of competitive advantage lies in distribution, brand identity, and the ability to manage AI-driven ‘agentic’ workflows securely.