Impact of Generative AI on Software Development

I’ve witnessed a dramatic rise in AI-assisted coding use in 2025 and was encouraged to use it at work through Cursor.

My experience

My anecdotal experience is that the autocomplete can be very good at guessing the next line of code you’re about to write, but that in agent mode where the LLM is performing larger more complex tasks on its own it’s prone to repeatedly getting things wrong and overengineering when it gets them right, and using it in this way is a net loss in productivity. However my experience is limited so things may improve as I get better at prompting and the LLM receives more feedback.

Research

• A 2025 study of experienced open-source developers found that “After completing the study, developers estimate that allowing AI reduced completion time by 20%. Surprisingly, we find that allowing AI actually increases completion time by 19%—AI tooling slowed developers down.”
• A 2024 study of 96 Google engineers found that AI assistance improved developer productivity by 21% on average, with the largest effect for more senior engineers and those who spend more time coding.
• A 2023 Github study on nearly a million Copilot users found that “users on average accept nearly 30% of the suggested code, leading to increased productivity. Furthermore…the acceptance rate rises over time and is particularly high among less experienced developers.” Though I wonder if the higher acceptance rates are due to less-experienced developers having lower standards.
• A 2023 study on users of Github Copilot found that “The treatment group, with access to the AI pair programmer, completed the task 55.8% faster than the control group” though it’s unclear how representative the task was of a work-like setting (since of course an AI can regurgitate FizzBuzz faster than a human)