Can Agentic Engineers Estimate Software Delivery in the Age of AI? (serendb.com)

In a thought-provoking exploration of software delivery in the AI era, it is suggested that agentic engineers may struggle to accurately estimate their engineering output when leveraging large language models (LLMs). Traditionally, software estimates relied on human intuition, past experiences, and the complexity of the codebase. However, with the advent of AI, especially LLMs, engineers face a paradox where these models tend to provide either overly optimistic or pessimistic time estimates, often failing to grasp the nuances of software development. This situation complicates planning and resource allocation, especially as projects grow in complexity. The significance of this discussion lies in the implications for the AI/ML community. While LLMs can expedite the programming process and assist in navigating complexities, their lack of human-like introspection and comprehension of tasks leads to unreliable estimates. As organizations increasingly integrate AI into their workflows, the challenge remains to develop methods that facilitate accurate and consistent software delivery estimates. The discourse raises questions about human capacity for software evaluations in complex projects and the evolving skills needed to effectively collaborate with AI in development, underscoring a vital area for future research and talent development in engineering.