In today's rapidly evolving tech landscape, where AI coding tools like ChatGPT, GitHub Copilot, and Cursor are taking center stage, it's crucial to gain a clear understanding of their true impact on productivity. Despite bold claims of 10x increases, our studio's experience, corroborated by industry data such as a 21% boost in developer productivity according to a Google study, points to a more moderate yet still significant gain in efficiency.

We're not in the strongly anti-AI camp. On the contrary, we're active proponents of its integration into our workflows. Our seasoned software engineers have noted a tangible increase in their output thanks to these tools. However, as a studio, we have a responsibility to consider the long-term implications of over-reliance on AI. Critical questions arise: Is it diminishing our own coding abilities? Is it depriving us of essential practice, thereby slowing down our learning and skill development?

While direct research into AI's impact on coding skills is still in its early stages, recent work from MIT on essay writing offers a valuable parallel. In that study, participants using ChatGPT "consistently underperformed" compared to those employing traditional methods, particularly when asked to quote from their own text. Furthermore, the non-AI group exhibited "more extensive brain network interactions," indicating deeper cognitive engagement.

While writing code differs from crafting essays, this correlation raises concerns. It's paramount to be mindful of how we leverage AI tools. Delegating the entire thought process to AI, as participants in the essay study seemingly did, could lead to a dependency that hinders skill development. At our studio, we encourage an approach where every engineer understands each line of generated code, turning it into a review and learning process rather than one of blind trust. Still, the question remains: Is reviewing as profound a learning experience as creating from scratch? Our current analysis suggests it is not.

Undeniably, AI tools provide significant time savings. Previously, finding an efficient implementation for a common function like debounce would involve searching platforms like StackOverflow. Now, tools like Cursor instantaneously provide these solutions. We see this as a judicious use of AI; not every aspect of development requires memorization of low-level implementations.

Moreover, for larger tasks like generating unit tests, AI can provide efficient starting points. While the output almost always requires tweaking, it considerably reduces the time spent on boilerplate creation and adapting existing files.

For our studio, where productivity and quality are key metrics, these AI advantages are invaluable. However, we also underscore the importance of continuous learning and a profound understanding of code. We firmly believe that in an ever-evolving job market, there will always be a high demand for engineers with deep coding expertise, regardless of LLM capabilities.

Thus, our approach is about finding the optimal balance. A complete rejection of AI would be counterproductive, especially given the growing prominence of "AI skills" in today's job market. However, we actively promote purposeful and strategic AI utilization, where engineers do not delegate core thinking processes to AI but rather leverage it to enhance efficiency and extend their capabilities.