Anthropic ran a controlled test where developers learning a new python library with an ai assistant…
anthropic ran a controlled test where developers learning a new python library with an ai assistant scored 17 percentage points lower on the final knowledge test than those who coded alone — and finished no faster. 52 participants, mostly juniors, and the assisted group averaged 50 percent.
the tool that writes the code can also stop you from learning the code.
Context
Anthropic's research post of 29 January 2026 reports a randomized controlled trial of 52 mostly junior software engineers, each using Python weekly for over a year and unfamiliar with the Trio async library. They completed two Trio features with or without a sidebar AI assistant, then took a quiz. The AI group averaged 50% versus 67% for hand-coding, a 17-point gap (Cohen's d 0.738, p 0.01), with the largest gap on debugging questions. The AI group finished about two minutes faster, not statistically significant. Heavy delegation scored under 40% on average, and users who asked for explanations retained more.
The note's numbers match. The gap is 17 percentage points, though Anthropic's prose writes 17% lower. Finished no faster is a fair reading of non-significance, not proof of equal speed. It covers one library, a short self-guided task, one platform, a quiz minutes after the task and a small sample, and Anthropic says AI would more likely help on repetitive or familiar tasks, so it is about learning an unfamiliar library, not coding ability in general. The paper's full methods were not read in full. The tool that writes the code can also stop you from learning it is the author's opinion, and the text supports can, not will.
Related work
- How AI Impacts Skill Formation (arXiv 2601.20245) ↗The paper, snippet only.
Watch next
- Replication with other libraries and experience levels.
Sources
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 18 September 2026 at 17:47 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →