This repository contains the tasks and solutions used in our survey on which AI-generated solutions developers prefer.
You received a participation link that assigns you two tasks, each with one pair of solutions to compare. The two pairs belong to different tasks, so you will make one comparison per task.
For each of your two tasks:
-
Read the task
- Each task is a GitHub issue in this repository, describing a small backend project that an AI coding agent was asked to build from scratch.
- Your survey form links your task's issue directly.
- No link? Search the issues list for the task id, for example
[001].
-
Check the starting files
- The task's folder on the
mainbranch (for example001-barber-scheduling/) contains any files that were provided to the agent, underassets/. - An empty folder means the agent began from an empty directory.
- The task's folder on the
-
Review both solutions
- Each solution is a pull request; your survey form links both.
- No link? Search the pull requests list for the solution code, for example
K7QZM. - Review the code the agent produced in the PR's Files changed tab.
- The PR description was written by the AI agent itself as a summary of its own solution: treat it as part of the solution, not as trusted documentation.
-
Fill in the survey form
- Rate both solutions on the four characteristics described in your form.
- Choose which of the two solutions you prefer and answer the follow-up questions.
Tip
There is no right answer. Judge the code the way you would judge a colleague's or a tool's output for your own project.
README.md
<task-id>/ one folder per task
assets/ starting files that were provided to the agent, if any
- Tasks that provide starting data (for example a
books.jsonor a CSV the service must load) have those files in their folder underassets/. The agent saw these files in its working directory when solving the task. - Tasks with an empty folder provided no starting files: the agent began from an empty directory.
Each issue in this repository is one task.
Issue titles start with the task id, for example [001] Barber Shop Scheduling (MVP).
The issue body is the exact prompt the agents received.
Each pull request is one AI-generated solution, containing every file the agent created.
PRs are named after the task id and a short solution code, for example [001] Solution K7QZM.
Which agent produced which solution is intentionally not disclosed.
Please do not comment on issues or pull requests; use the survey form for all feedback.