INQUIRING LINE

Do experienced programmers really gain most from AI coding tools, given one trial where veterans took 19% longer?

Why do experienced developers benefit more from AI coding assistance?

This explores whether experience changes how much developers gain from AI coding tools — and the corpus pushes back on the premise, because the strongest evidence here suggests experienced people often gain less, or even lose ground.


This explores whether experienced developers get more out of AI coding tools than newcomers do. The short answer from this collection: the premise doesn't hold up well. The most direct test points the other way. In a randomized trial, experienced open-source developers worked on real tasks in codebases they knew deeply. With early-2025 AI tools they took 19% longer, even though they had predicted a 24% speedup and still felt faster afterward Do AI coding tools actually speed up experienced developers?. Part of the explanation is the expertise itself. When you already know a mature codebase well, unreliable suggestions can cost more in checking and fixing than they save.

The same pattern shows up outside software. In a study of more than 5,000 customer-support agents, AI assistance raised productivity 15% on average. Most of that gain went to the least experienced workers. The most experienced agents got only small speed gains and saw their quality slip slightly Does AI assistance help less experienced workers most?. A plausible reading is that AI spreads the know-how of top performers to everyone else. That helps novices a lot and gives experts little they didn't already have, plus some noise. Not every result is negative: a trial of 96 Google engineers found AI features cut time on a complex task by about 21%. The uncertainty around that number was wide, though, and the study didn't break the result down by experience Do AI coding features actually speed up engineer productivity?.

So where does the idea that experts benefit more come from? The best version of it is about oversight, not speed. Anthropic surveyed its own engineers. They reported about 50% productivity gains, yet most said they could fully hand off only 0–20% of their work. They also worried that leaning on AI for routine coding would wear down the hands-on skill they need to catch its mistakes Does AI assistance erode the skills needed to oversee it?. Experience may matter less for getting more out of AI than for knowing when the AI is wrong. That turns the question around: AI may help experienced developers safely, while novices get the speedup without the judgment to check it.

The less obvious finding is that how you use the tool may matter more than how experienced you are. In a trial of developers learning a new library, the passive ways of using AI produced quiz scores of 24–39%. The engaged ways, where people worked to understand what the AI produced, reached 65–86% Does AI assistance actually harm the way developers learn?. A separate study found that patterns in people's conversations with coding agents predicted results beyond what their prior skill explained. However, those patterns weren't stable enough to count as teachable skills Can conversation patterns predict coding outcomes better than prior skill?. There's also a hidden cost that may hit experts hardest: even correct AI suggestions can break a person's focus and force them to rebuild their train of thought Does AI assistance always help reasoning or does it carry hidden costs?. Experts working in deep focus have the most to lose from those interruptions.

One caveat: no study in this collection directly compares experienced and novice developers on the same coding tasks. The 'novices gain more' pattern comes most clearly from customer support, and the 'experts slow down' result comes from a small sample of 16 developers. Treat both as strong signals rather than settled answers.


Sources 7 notes

Do AI coding tools actually speed up experienced developers?

A randomized controlled trial of 16 developers on 246 real tasks found completion times increased 19%, despite developers forecasting a 24% speedup beforehand. Experts in economics and ML also overestimated gains; slowdown factors included over-optimism, low AI reliability, and developers' deep familiarity with mature codebases.

Does AI assistance help less experienced workers most?

A study of 5,172 support agents at a Fortune 500 firm found a 15% average productivity gain from AI assistance, with gains concentrated among less experienced workers who improved both speed and quality. The most experienced agents saw small speed gains but slight quality declines.

Do AI coding features actually speed up engineer productivity?

A randomized trial of 96 Google engineers found AI Code Completion, Smart Paste, and Natural Language to Code shortened time on a complex task by roughly 21%, though the confidence interval was wide and statistical significance depended on model specification.

Does AI assistance erode the skills needed to oversee it?

Anthropic's 132-person survey found 50% self-reported productivity gains and 67% more merged pull requests, yet most engineers can only fully delegate 0-20% of work. Employees fear that relying on Claude for routine tasks erodes the hands-on coding practice needed to catch its errors.

Does AI assistance actually harm the way developers learn?

A randomized trial of developers learning new libraries showed AI use degraded conceptual understanding and debugging ability. Six interaction patterns emerged: three low-engagement patterns produced quiz scores of 24-39%, while three high-engagement patterns with active comprehension steps achieved 65-86%, suggesting the mechanism matters more than tool presence.

Show all 7 sources
Can conversation patterns predict coding outcomes better than prior skill?

Machine learning identified interpretable traits from coding-agent conversations that explained outcomes beyond prior achievement. However, these traits lacked the stability and transferability required to qualify as learnable human-AI collaboration skills.

Does AI assistance always help reasoning or does it carry hidden costs?

Well-intentioned AI suggestions can damage reasoning performance by severing cognitive immersion, forcing users to rebuild focus before continuing. Evaluation must measure flow preservation across entire tasks, not just local suggestion accuracy.

Papers this line draws on 8

The research behind the notes this line reads — ranked by how closely each paper relates.