The 27 million developers on the planet are the darlings of digitalization. While the need to code tomorrow’s software and maintain yesterday’s is exploding, the educational sphere is not keeping up. And this is despite a 42% increase in the number of training courses in the field in ten years, according to Evans Data Corporation. Fortunately, this industry is experiencing a revolution: the arrival of programming assistants supported by artificial intelligence.
GitHub, the world’s largest source code hosting service, founded in 2008 and acquired ten years later by Microsoft for $7.5 billion, is the essential tool for developers. An estimated 93% of them use it. In June 2022, five months before the launch of ChatGPT, GitHub announced the launch of its Copilot assistant, based on Codex, an OpenAI model specializing in code generation. It now has 1.5 million paying users and is said to generate around $300 million in revenue. The tool has written almost half – 46% to be exact – of the code of these users and has increased their productivity by 55%.
A third of working time saved
These are not just numbers bandied about for advertising purposes. A scientific study sought to evaluate the efficiency gains related to the use of Copilot, identifying 15 software development tasks. The results indicate significant reductions in working time in developers’ missions. Up to 50% in code documentation and auto-completion. Between 30 and 40% in repetitive coding operations, unit test generation, debugging and pair programming. Overall, the authors predict a reduction of 33 to 36% in the time required for coding activities.
Microsoft is not alone in this arena. Amazon CEO Andy Jassy revealed that their software development assistant, Amazon Q, has saved the company 4,500 years of development time. That’s an estimated annualized gain of $260 million. The Cursor code editor, which integrates one of the language models of the American company Anthropic, received high praise from one of the co-founders of OpenAI, Andrej Karpathy, on the social network X: “Most of my programming now consists of writing in English, asking for and reviewing and editing the differences generated, and doing a little half-coding.”
About ten startups are currently working on software development assistants. And Copilot’s economic success is whet investors’ appetite for these companies. In August, Anysphere, the startup that developed Cursor, raised more than $60 million from California-based Andreessen Horowitz. It had already raised $11 million last year from OpenAI’s venture capital fund and former GitHub CEO Nat Friedman.
Massive fundraising
Notable competitors in the space include Augment, Replit, Cognition, Poolside, Supermaven, and Magic. The latter, which employs just 20 people, is reportedly in talks to raise $200 million, while Poolside is reportedly seeking $450 million. In April, Augment raised $252 million, while Cognition, whose tool is called Devin, secured $175 million. The sector is likely to be a winner-take-all market, with no single-leader capture effect. Development strategies differ. Some, like Magic, Poolside, Augment, and Supermaven, pre-train their own large AI models, focusing on context and latency, among other things. Others, like Cursor, are model-agnostic and focus on developer experience, interface, and workflows.
These upcoming innovations are excellent news. Given the current shortage, developers are unlikely to be replaced. Our children can embrace the profession without worrying about their employability, even if the productivity gap between a very good developer and an average one could widen. The other good news? The growing stock of code written in old languages like Cobol, which is becoming obsolete even though it runs essential systems in our societies – from hospitals to banks – will be able to be updated much more easily. We can breathe easier.
Robin Rivaton is CEO of Stonal and member of the scientific council of the Foundation for Political Innovation (Fondapol)
.