Why the productivity gains from AI are still small.

Hacker News is a community-driven platform for sharing and discussing technology news, startups, and programming-related topics. Through user submissions and comments, Hacker News offers insights into emerging technology trends, industry developments, and entrepreneurial ventures. Readers can participate in discussions, share their insights, and stay informed about the latest advancements in technology and innovation.

Hacker News

AI tools accelerate code writing significantly, but coding is only a fraction of a developer's day. Using time-breakdown tables for both senior and junior developers, the analysis shows AI yields roughly 15% efficiency gains for seniors and 25% for juniors — far below the hype. Senior engineers spend most of their time on design, architecture, meetings, and mentoring, areas AI hasn't meaningfully improved. Counterintuitively, junior developers benefit more from AI than seniors, making the common leadership stance of 'AI replaces juniors' backwards. Dramatic productivity gains remain elusive until AI improves at the non-coding parts of the job.

The AI productivity gap

We’re all trying to skate around the truth that LLMs <a href="https://larsfaye.com/articles/agentic-coding-is-a-trap" target="_blank" rel="noopener nofollow">accelerated the wrong parts</a> of the development cycle. At least, if your goals are to maintain quality and expertise. If speed and velocity is all you really care about, then sure, they’re just incredible. But if you even slightly care about all the other elements that go into good coding and development practices (including the growth &amp; retention of your own skills), the main benefits of LLMs begin to diminish in their usefulness.
Of course, you can go full on Robert “Uncle Bob” Martin and stop looking at the code entirely, but when you dig into what he actually does, the amount of guardrails he’s constructed is non-trivial, and he’s admitted that most of his work is greenfield personal apps. And let’s not forget that he’s been coding for SIXTY years, and he <a href="https://www.youtube.com/watch?v=RxxxGkFIUJ0&amp;t=1356s" target="_blank" rel="noopener nofollow">also says</a> that junior developers shouldn’t even touch AI agents for three full years (and should spend those years writing code).

Ai increase productivity if used efficiently as a tool, Ai will help us to save more time. If a developer who uses AI as a tool he may exceeds other developers only if he who learns from it.

Juniors seem to get the biggest boost because AI helps them overcome blockers faster.

The 15% for seniors and 25% for juniors matches what actually happens once you count the parts of the day that are not typing code. Review, design, and coordination do not compress just because generation got faster, so the headline speedups never translate cleanly. Did the breakdown account for the extra review load that agent-written code adds back onto seniors?

Speed is good if you know what are you doing and can recognize when the output is wrong but most people don’t know it as they outsource everything to the ai. Person has to spend time typing the code and to build understanding of the code. Than he will recognize bad and good code parts generated by ai.