New research from Stack Overflow has revealed a troubling relationship between IT teams and AI coding assistants.
Despite high adoption rates, concerns about productivity, complexity and trust persist, indicating that many teams are simply using emerging generative AI tools out of curiosity, or to blindly stay ahead of the curve.
A poll of over 1,700 developers showed that around three in four (76%) are using, or plan to use, AI code assistants. These tools have become popular among academic researchers, AI developers, front end developers, mobile developers and data scientists.
Developers aren’t benefitting from generative AI
Two in five (38%) expressed concerns that AI tools can provide inaccurate information at least half of the time, which aligns with a Purdue University study which claimed that more than half (52%) of the responses that ChatGPT produced are incorrect. This inaccuracy, combined with increased complexity and trust issues, becomes even more pronounced as team size grows.
Despite these issues, the majority (95%) of developers report that AI tools can enhance their productivity to varying degrees. To coincide with this, the perception of productivity has become a grey area, with three-quarters (74%) unsure how their organization measures productivity.
Among the most highly regarded tools are Codeium, GitHub Copilot and ChatGPT, with developers rating their ease of use strongly. However, struggles with context, complexity and niche scenarios proved to be limitations of all AI coding assistants.
On the whole, the findings indicate that while AI tools may not always deliver on productivity promises directly, they enhance job satisfaction and quality of work life, which is likely to positive impact productivity indirectly.
