If you work in the digital economy, you know the feeling: the tools you mastered three years ago are already starting to feel obsolete. The landscape of programming technologies—languages, libraries, frameworks—moves at an ever-faster pace. But is this seemingly chaotic movement random, or are there hidden laws that govern which technologies rise to the top and which ones fade into obscurity?

In a new paper, published in Royal Society Interface, we search for these laws in the world’s largest archive of programming knowledge: Stack Overflow. Our research analyses over a decade of online trace data to understand the innovation dynamics of the software world. What we found provides an empirical window into how innovation works—and it shows that “creative destruction” (one of the most famous theories in economics) seems to govern the evolution of digital technologies, such as it governs the renewal of economies.

The ebb and flow of the developer community

We analysed over 22 million questions asked by more than four million users on Stack Overflow between 2008 and 2022. Rather than just counting which tags were most popular, we constructed so-called correlation networks from the data.

In other words, we didn’t just look into the popularity of certain programming technologies, such as Python or React, over time. Instead, we were interested in finding out which technologies were “rising and falling together”. If two technologies consistently grew in usage at the same time, they share a positive link. If one rose while another fell, they share a negative link—a sign of competition between the technologies.

When we mapped this out, two distinct worlds emerged:

1. Core Computing Facilities: This cluster includes the “heavy lifting” infrastructure—operating systems, databases, and servers.
2. Application Development: This cluster includes modern toolkits—web frameworks, data science libraries, and machine learning tools.

Creative destruction in action

The most exciting finding wasn’t just what was popular, but how it became popular. Our data supports Joseph Schumpeter’s famous theory of creative destruction. Schumpeter argued that innovation is not just about adding new things; it is a “process of industrial mutation… that incessantly destroys the old one, incessantly creating a new one”. Our model revealed a formula for success in the programming world. The technologies that became “winners” (the top 100 most used tags) shared a specific pattern of connections:

1. They play nice with other new kids: Successful new techs had strong positive ties to other emerging technologies. They didn’t grow in isolation; they grew as part of a new ecosystem.
2. They compete with the giants: Crucially, successful new techs often had strong negative ties to established, popular technologies. They didn’t just supplement the old guard; they replaced it.

For example, we saw this dynamic clearly in the rise of Swift (which replaced Objective-C) and Kotlin (which challenged Java/C# in mobile dev). The data suggests that to really succeed, a new technology often needs to be a “substitute” that renders an older tool unnecessary, rather than just a “complement” that works alongside it.

Figure 1 shows distinct patterns regarding the rise and decline of old and new technologies, coloured distinctly.

(A) The rise and fall of programming technologies does not happen in isolation. Instead, the dynamics of programming technologies form complex networks with two distinct clusters emerging that shift over time.

(B) In both clusters – application development and core computing – the share of declining and rising technologies shifts over time: often the rise of successful technologies happens at the cost of previously established technologies; a process described as “creative destruction”.

Why this matters now: AI, skills, and economic complexity

These findings have several implications that connect directly to current debates about programming and digital technologies.

1. Early warning signals of technological shifts
   Because Stack Overflow captures behaviour from millions of developers, changes in these correlation networks could provide early indicators of major shifts in programming paradigms – much earlier than official statistics or enterprise adoption data.
2. Labour markets and skills
   Recent work shows that developer task profiles inferred from online platforms predict the skills demanded in job adverts and even salaries. Our results suggest that we can go further and identify which skills are likely to gain or lose relevance as technologies compete and recombine.
3. AI-assisted programming and the next wave of creative destruction
   The study notes that the rapid uptake of generative AI tools, such as ChatGPT and Copilot, is already changing how developers seek help and share knowledge, reducing activity on Q&A platforms like Stack Overflow. If AI accelerates experimentation and lowers switching costs between tools, we may see faster, more volatile cycles of technological rise and decline — with significant consequences for training, productivity, and inequality between regions and firms who might struggle to adapt fast enough.
4. Economic complexity and national capabilities
   Programming technologies are part of the “software complexity” of countries – the richness of their digital capabilities. Understanding how digital tools evolve can help policymakers anticipate which regions are well-positioned to benefit from new technological clusters and which risk being left with declining stacks.

From digital trace data to innovation policy

Our study shows that a platform built initially for debugging code has become an invaluable dataset for understanding how digital innovation actually unfolds.  For researchers, correlation networks of developer activity offer a scalable way to model technological recombination and competition, complementing traditional patent and trade data. For journalists and policymakers, they provide a way to watch the software world’s creative destruction unfold in almost real time – revealing not just which technologies are popular, but how the next generation of tools is quietly reshaping the foundations beneath them.

You can read the full paper, “The innovation dynamics of programming technologies“, authored by Conrad Bouchers and Dr Fabian Braesemann, published in the Journal of the Royal Society Interface.

About the authors

Dr. Fabian Braesemann is a Departmental Research Lecturer in AI & Work at the Oxford Internet Institute and an Associate Faculty member at the Complexity Science Hub. Fabian is also an Associated Researcher at the Einstein Center Digital Future in Berlin and Managing Director and Co-Founder of the DWG Data Science Company.

Conrad Borchers is a PhD candidate at the Human-Computer Interaction Institute (HCII) at Carnegie Mellon University, School of Computer Science where I am advised by Vincent Aleven and Ken Koedinger. He is an OII alumnus and former MSc student.