Enterprise AI Pilots Stall Due to Poor Systems Design, Not Models

Enterprise deployments of agentic coding AI systems have underperformed in most cases, according to recent reports. The issue is not with the AI models themselves, but rather with the environment in which they operate. Experts point to the lack of proper systems design as the primary limiting factor. Dhyey Mavani, a leading researcher in the field, notes that the shift from assistive coding tools to agentic workflows has been rapid in the past year. "Research has begun to formalize what agentic behavior means in practice: the ability to reason across design, testing, execution, and validation rather than generate isolated snippets," Mavani said in an interview. This new approach has shown significant improvements in outcomes for large, interdependent codebases. Dynamic action re-sampling, a technique developed by researchers, has demonstrated that allowing agents to branch, reconsider, and revise their own decisions can lead to better results. Providers like GitHub are now building dedicated agent orchestration platforms to support this new type of coding. The problem lies in the context surrounding the code being changed, including its structure, history, and intent. This requires a deeper understanding of the systems in which the AI agents operate. "Enterprises are facing a systems design problem: they have not yet engineered the environment these agents operate in," Mavani explained. The industry has been moving towards more autonomous coding, with the goal of reducing the need for manual intervention. However, the complexity of large codebases and the need for context-aware decision-making have proven to be significant challenges. As a result, many enterprise deployments have underperformed, failing to meet expectations. Despite these challenges, researchers and developers remain optimistic about the potential of agentic coding. "We're at the beginning of a new frontier in software engineering," Mavani said. "With the right systems design and environment, we can unlock the full potential of AI agents in coding." The current status of agentic coding is that it is still in its early stages, with many companies experimenting with different approaches. Next developments are expected to focus on improving systems design and creating more context-aware AI agents. As the field continues to evolve, it is likely that we will see significant improvements in the performance of agentic coding AI systems.