# How AI assistance impacts the formation of coding skills

Research shows AI helps people do parts of their job faster.

| | |
| --- | --- |
| Status | 9 MIN READ |
| Source | web:www.anthropic.com/research/AI-assistance-coding-skills |
| Original | https://www.anthropic.com/research/AI-assistance-coding-skills |
| Workspace | Showcase |
| Tags | web, document |

## What it says

**A 9-minute document, structured into 6 sections.**

2,175 words across 6 sections, roughly 9 minutes of reading on www.anthropic.com.

Research shows AI helps people do parts of their job faster.

**Diagram: The document's own headings, in order, sized by how much sits under each. Longest sections are marked.**

- How AI assistance impacts the for… (381 words) [accent]
- Study design (179 words) [neutral]
- Evaluation design (189 words) [neutral]
- Results (94 words) [neutral]
- Qualitative analysis: AI interact… (621 words) [accent]
- Conclusion (684 words) [accent]
- How AI assistance impacts the for… -> Study design
- Study design -> Evaluation design
- Evaluation design -> Results
- Results -> Qualitative analysis: AI interact…
- Qualitative analysis: AI interact… -> Conclusion

## Key claims

**What each section of the document is about.**

Taken from the headings and the opening line beneath each. The document's own structure, not an interpretation of it.

- **How AI assistance impacts the formation of coding skills.** Jan 29, 2026 Read the paper Research shows AI helps people do parts of their job faster.
- **Study design.** We recruited 52 (mostly junior) software engineers, each of whom had been using Python at least once a week for over a year.
- **Evaluation design.** In our evaluation design, we drew on research in computer science education to identify four types of questions commonly used to assess mastery of coding skills: - Debugging : The ability to identify
- **Results.** On average, participants in the AI group finished about two minutes faster, although the difference was not statistically significant.
- **Qualitative analysis: AI interaction modes.** We were particularly interested in understanding how participants went about completing the tasks we designed.
- **Conclusion.** Our results suggest that incorporating AI aggressively into the workplace, particularly with respect to software engineering, comes with trade-offs.

## Receipts

**Quoted from the page, with the heading each line sits under.**

Every line below is verbatim, so the page can be checked against this brief rather than trusted over it.

- **Claim.** The section "How AI assistance impacts the formation of coding skills" opens on this.
  - Evidence: Jan 29, 2026 Read the paper Research shows AI helps people do parts of their job faster.
  - Where: www.anthropic.com — How AI assistance impacts the formation of coding skills
- **Claim.** The section "Study design" opens on this.
  - Evidence: We recruited 52 (mostly junior) software engineers, each of whom had been using Python at least once a week for over a year.
  - Where: www.anthropic.com — Study design
- **Claim.** The section "Evaluation design" opens on this.
  - Evidence: In our evaluation design, we drew on research in computer science education to identify four types of questions commonly used to assess mastery of coding skills: - Debugging : The ability to identify
  - Where: www.anthropic.com — Evaluation design
- **Claim.** The section "Results" opens on this.
  - Evidence: On average, participants in the AI group finished about two minutes faster, although the difference was not statistically significant.
  - Where: www.anthropic.com — Results
- **Claim.** The section "Qualitative analysis: AI interaction modes" opens on this.
  - Evidence: We were particularly interested in understanding how participants went about completing the tasks we designed.
  - Where: www.anthropic.com — Qualitative analysis: AI interaction modes
- **Claim.** The section "Conclusion" opens on this.
  - Evidence: Our results suggest that incorporating AI aggressively into the workplace, particularly with respect to software engineering, comes with trade-offs.
  - Where: www.anthropic.com — Conclusion

## Gaps

**What this brief did not do.**

Worth stating plainly, because a summary that hides its own limits is the thing that builds the debt rather than paying it down.

- **Nothing here was understood** (high)
  - The sections, quotes, and counts are mechanical extractions. No model read the argument, so nothing in this brief tells you whether the page is right, or even coherent.
  - Mitigation: Read the sections marked longest. They are where the reasoning usually is.

## What this brief could not check

- No language model read this page. The outline, the quotes, and the counts are mechanical, so this brief tells you how the document is arranged and nothing about whether its argument holds.