Skip to content
Trending storyDeveloping

Anthropic test: multi-step workflow finds far more planted bugs than a single agent

1 article1 sourcesince Oct 9Last article 2h ago ·

Overview

AISummary of 1 article

Anthropic reports that in a bug-detection experiment on a 116k-line codebase with 70 planted bugs, a workflow found 66 bugs in each of three runs, while a single agent found 14, 15 and 27.

The result is posted by ClaudeDevs on X; the report describes only the planted-bug setup and counts, without details on how the workflow or agent was built or how the bugs were planted.

Written by AI from the articles below · updated Oct 9, 12:27 PM ET

Check the sources:

Latest developmentOct 9, 12:12 PM ET
Anthropic workflow finds 66 planted bugs; single agent finds 14 to 27

Article timeline

The articles in this story. Times are ET.

Oct 9
  1. ClaudeDevs
    Anthropic workflow finds 66 planted bugs; single agent finds 14 to 27

    AIAnthropic planted 70 bugs in a 116k-line codebase and tested two approaches over three runs. A single agent found 14, 15 and 27 bugs, while a workflow found 66 bugs in each run.

Heat trend

Not enough continuous observations to show a trend yet.