The AI Jockey · 2026-10-13

Does AI coding agents efficiency actually increase software output?

According to a Harvard study of over 700 firms, AI coding agents do not significantly increase software output despite generating 30% more code. The gains are offset by a 49% increase in the time required for human code review, creating a production bottleneck.

Key points

Lines of code generated increased by 30% with AI agents.
Pull requests increased by 23% on average.
Code review time ballooned by 49% after AI introduction.
No statistically significant change in Jira issue resolution rates.

Sources

AI coding agents generate more code, but not more software

FAQ

Why isn't AI making software development faster?

The study found that human reviewers spend significantly more time checking AI-generated code, which leads to more comments and revisions, neutralizing the speed of code generation.

Video transcript (Hinglish)

Kya AI coding agents efficiency badha rahe hain ya sirf kachra generate kar rahe hain? Chalo batata hoon kaise, lekin pehle AI Jockey ko follow kar lena. Harvard researchers ki ek nayi study ne coding agents ka sach samne laya hai. Data ke mutabik AI agents ki wajah se total lines of code tees percent badh gaya hai. Lekin itna sara extra code hone ke bawajood software output bilkul nahi badha. Asal mein human code review ab ek bada bottleneck ban chuka hai. Review process ka time lagbhag pachas percent tak lamba ho gaya hai. Researchers ne paya ki Jira features resolve hone ki speed mein koi sudhaar nahi hua. AI coding agents efficiency badhane ke bajaye reviewers par bojh badha rahe hain. Ye report software engineers aur tech leads ke liye ek bada reality check hai. Kya aapko lagta hai AI coding agents kaam ke hain? Comment karo BOTTLENECK, main bhej deta hoon.

Made with an AI avatar and AI voice of the creator, with consent. Want the links in your DMs? Comment "BOTTLENECK" on the reel on Instagram.