All this investment in AI, why aren't we shipping faster? (medium.com)

A recent Stanford study highlights that over a quarter of a trillion dollars has been invested in AI, prompting the question of why such innovation isn't accelerating software delivery. The study reveals a paradox where current AI tools enhance specific tasks—like code authoring—rather than streamlining the overall software delivery system. While technologies such as GitHub Copilot and Claude Code expedite coding, they don't address the significant bottlenecks sidetracking enterprise software deployment. Compliance, security, and testing issues—rooted in archaic organizational workflows—often prolong the time it takes to move code from development to production. This disconnect between AI capabilities and enterprise constraints underscores a critical challenge for the AI/ML community: merely increasing coding capacity can lead to higher work-in-progress (WIP) issues, delaying feedback and complicating integration. As organizations adopt AI, they often experience initial productivity dips due to learning curves and the need to recalibrate existing workflows. The study advocates for focusing on enhancing processes downstream of coding, such as automating code reviews and deployment tasks. Emphasizing AI's role in reducing coordination costs and streamlining operational flows could unlock substantial improvements in software delivery efficiency, addressing long-standing challenges in the industry.