Why aren't we shipping faster now?

In a recent discussion at the UC Irvine software engineering seminar on October 30, 2026, experts debated whether the rapid adoption of artificial intelligence in software development is leading to faster delivery times. The panel included researchers from various institutions, all of whom agreed that while AI tools have significantly improved coding productivity, they have not yet translated into a noticeable increase in software shipping speed.

One of the key findings presented was a study titled "Writing Code vs. Shipping Code: Productivity Effects Across Generations of AI Coding Tools." Conducted by Demirer, Musolff, and Yang, this research, now available as NBER Working Paper 35275, compared the productivity gains of early AI coding tools against later versions. The findings suggested that while AI can enhance individual coding efficiency, the subsequent impact on the overall speed of software delivery remains limited.

Another notable contribution came from Jensen Huang's keynote address at Computex in May 2023, where he highlighted the potential of AI to unlock new possibilities for developers. Huang's insights were later echoed in Yang, He, Cheng, Butler, Murphy-Hill, Badea, and Bird's arXiv paper, "When Should a Software Factory Be Lights-Off?

Benchmarking Code Review Agents on Bug Detection." This work explored the conditions under which developers might benefit most from AI-driven code review agents, suggesting that the integration of AI into the software development process requires careful consideration of various factors.

The panel also delved into the challenges of AI autonomy in software development. Choudhuri, Bird, Badea, Gerosa, Sarma's paper, "You Shall Not Pass! Where and Why Developers Draw the Line on AI Autonomy," published on arXiv as 2607.00533, examined the boundaries that developers are willing to set when it comes to delegating decision-making to AI systems.

The study identified specific scenarios where developers prefer to maintain human oversight over AI-generated code, indicating that fully automated software creation processes are not yet a practical reality.

Miller, Choudhuri, Ulloa, Haniyur, DeLine, Storey, Murphy-Hill, Bird, and Butler's ICSE 2026 paper, "Maybe We Need Some More Examples": Individual and Team Drivers of Developer GenAI Tool Use, highlighted the diverse motivations behind the adoption of Generative AI (GenAI) tools. Published on arXiv as 2507.21280, this research pointed to a mix of individual curiosity, team collaboration incentives, and project-specific needs as drivers for developers' willingness to incorporate AI into their workflows.

Despite these insights, the consensus among the seminar participants was clear: while AI has undoubtedly improved coding efficiency, the gap between coding productivity and actual software shipping speed remains substantial. The panelists agreed that further research and development are necessary to bridge this AI-activity gap and realize the full potential of AI in accelerating software delivery.