Developers Are Writing More Code but Shipping Fewer Products
In early 2025, software company leaders began reporting a curious trend. Engineering teams were producing code at unprecedented rates thanks to AI coding assistants such as GitHub Copilot, Cursor, and Claude. Pull requests increased. Code commits increased. Development activity surged.
Yet many executives struggled to identify a corresponding increase in completed software products, customer adoption, or business value. One technology executive summarized the situation to me perfectly. “We’re producing more code than ever. I’m not sure we’re producing more software.”
At Sonatafy, we call this the AI Productivity Paradox. Artificial intelligence is dramatically increasing coding activity while delivering far smaller gains in shipped software products.
For years, I’ve seen software leaders have measured productivity through engineering outputs. Lines of code. Story points. Features completed. Development velocity. AI excels at all of those metrics.
Developers can generate boilerplate code in seconds. Documentation can be drafted automatically. Unit tests can be created almost instantly. Entire application components can be scaffolded through natural-language prompts. The result is an explosion of code generation.
Yet software products don’t succeed because code exists. They succeed because customers use them, businesses support them, and organizations can maintain them over time. Those challenges remain largely human. The evidence is beginning to support this distinction.
Productivity Gains?
GitHub’s 2024 Developer Survey found that developers using AI coding tools reported significant gains in productivity and efficiency. Participants said AI helped them complete repetitive tasks faster and reduced time spent searching for information. The survey showed that developers overwhelmingly believed AI was making them more productive. At first glance, that sounds like an unqualified success story.
The Stanford University 2025 AI Index Report painted a more nuanced picture. While AI coding assistants continued to improve software development speed, Stanford researchers noted that organizations were still struggling to translate AI-driven productivity gains into measurable business outcomes. The report highlighted a growing gap between task-level productivity improvements and organization-wide performance gains. In other words, developers were completing coding tasks faster than companies were delivering business value.
The 2025 Microsoft Work Trend Index revealed a similar pattern. Researchers found that employees increasingly relied on AI to accelerate individual work activities. Yet many organizations continued to face bottlenecks involving decision making, collaboration, approval processes, and organizational complexity. Faster execution at the individual level did not automatically eliminate constraints elsewhere in the business.
Writing Code is Only one Stage of Software Delivery
Before a product reaches customers, teams need to gather requirements, align stakeholders, perform testing, address security concerns, manage deployment pipelines, train users, support customers, and continuously improve the product based on feedback. AI helps with some of those activities. But it doesn’t eliminate them.
Research from Atlassian’s 2024 State of Teams report found that software developers spend substantial portions of their week on communication, coordination, planning, meetings, and information gathering. Coding represents only part of the software development process. Accelerating code creation therefore improves only one segment of a much larger workflow.
This helps explain why many companies report impressive AI adoption statistics while struggling to identify dramatic improvements in software delivery.
A 2025 McKinsey Global Survey on AI found that organizations were expanding AI deployments across business functions, yet relatively few had achieved significant enterprise-wide financial impact. Most reported gains remained localized to specific teams or individual workflows rather than transforming overall organizational performance.
That finding shouldn’t surprise us. Businesses don’t buy code. And customers don’t purchase pull requests. They purchase solutions to problems.
Many executives assume that if AI allows developers to generate twice as much code, software products should arrive twice as fast. Reality rarely works that way. That’s because the constraint was never purely technical. The constraint was organizational.
The Metrics That Matter
At Sonatafy, we believe that the companies that benefit most from AI will stop measuring success by the volume of code generated and start measuring outcomes customers actually experience.
Did adoption increase? Did revenue grow? Did deployment cycles shorten? Did customer satisfaction improve? Did business objectives get achieved?
AI is becoming an extraordinary amplifier for software developers. It is reducing friction, accelerating experimentation, and making coding more accessible than ever before. What it hasn’t done is eliminate the hard work of building products people actually want. That remains a human challenge.
The AI Paradox is that we now possess technology capable of generating more software code than any previous generation of developers could imagine. Yet the organizations that win won’t be the ones producing the most code. They’ll be the ones producing the most value.
WANT TO LEARN MORE? DOWNLOAD OUR FREE REPORT:
