← Cases
ChinaFinance and insurance2023 – ongoing

Ant Group

CodeFuse internal software-development assistant

Ant's CodeFuse users produced 55% more code in a matched comparison

WorkingCompany's own claim

Problem

Ant tested whether a coding language model could increase programmer output.

Approach

Ant introduced CodeFuse in September 2023 to selected programmers before wider release. Researchers compared users with similar employees who lacked access during the initial trial.

Results

55%increase in code-line output in the matched comparison reported by BIS and Ant-affiliated researchersCompany's own claim[1]

Takeaway

Statistically significant gains were concentrated among junior staff. Output was measured in lines of code, and an Ant employee coauthored the study.

Categories

Business function
IT and R&D
Innovation type
Process innovation
AI technology
Generative AI
Evidence
Company's own claim

Sources

  1. Generative AI and labour productivity: a field experiment on coding
    Bank for International Settlements4 Sep 2024Research paper or report

Source summaries

[1]Bank for International Settlements4 Sep 2024

The researchers estimated a 55% code-output increase after CodeFuse became available. Significant gains were concentrated among junior programmers; one author worked for Ant, which the authors said did not influence the content.

Full details are in the original sources.

Added 1 Nov 2026 · Updated 1 Nov 2026
Open case page ↗