Can a Pet Improve Code Review Moods? A Dev Experiment

What a code review pet can and cannot change

A code review pet does not read your diff and it will never tell you a test is missing. What it does is narrower: it sits on your desktop while you read somebody else's work, and it gives the flat, slightly tense parts of review somewhere to go. We wanted to know whether that is worth anything, so we ran a developer happiness experiment on one team for two weeks.

The short version: review threads got shorter and the wording in them got less sharp. Nobody became a better engineer. If you are hoping for something that makes code review enjoyable, this is not that.

How the experiment was set up

Seven engineers on one product team, all reviewing in the same repo, all running Codex. Each person installed a pixel pet generated from a photo they picked themselves, and we agreed on one rule: nobody changes the review process. Same checklist, same turnaround target, same reviewers.

For every pull request we logged three things: how many comments it drew, how many of those comments were about style rather than substance, and a one to five mood score the author recorded right after reading the feedback. The third number is self reported and soft. The first two are not.

• Baseline week: no pets installed, 41 pull requests
• Test week: pets installed, 38 pull requests
• Identical review checklist and identical reviewers in both weeks
• Mood score filled in within ten minutes of opening the feedback

What the numbers showed

Comments per pull request went from 6.4 to 5.1. Style-only comments, the ones that open with "nit:" and cost an extra round trip, fell from 2.3 to 1.2 per request. The self reported mood score moved from 3.1 to 3.6, which is a real shift but small enough that we would not defend it on its own.

Turnaround time did not move. Neither did the number of defects found. Whatever changed, it changed in how review feels rather than in what review catches.

• Comments per pull request: 6.4 at baseline, 5.1 with pets
• Style-only comments: 2.3 at baseline, 1.2 with pets
• Self reported mood after reading feedback: 3.1 to 3.6 out of 5
• Time to first review: unchanged at roughly four hours in both weeks
• Two of seven engineers reported no difference at all

Where a review mood pet helps and where it does nothing

The effect showed up in one specific moment: right after you open a review that has twenty comments on it. That is where people write the reply they later regret. A pet does not fix the review. It puts something calm and predictable in the corner of the screen while you decide what to say.

It did nothing for the engineers who already enjoyed review, and nothing for the two who found the pet distracting in the first three days and moved it off screen.

• Helps when feedback is long and you need a beat before replying
• Helps on the days you have been reading other people's code for four hours
• Does nothing when the review itself is genuinely unclear
• Backfires if the pet sits on top of your diff, so park it away from the editor

Running your own developer happiness experiment

Two weeks is long enough to see something and short enough that nobody complains about it. Pick one repo, keep the process identical, and write the numbers down at the end of each week instead of guessing at the end.

• Generate one pet per person so nobody shares a companion
• Count comments per pull request before you install anything
• Keep the review checklist fixed across both weeks
• Write the mood score down immediately, not from memory on Friday
• Stop if anyone finds the pet distracting, because that result counts too

Try it on your next review cycle

Start at /, generate a pet from any photo you like, and read more about why these companions stick at /blog/why-developers-love-desktop-companions. Plan limits are listed on /pricing. If your pet never appears on the desktop, the checklist at /blog/codex-pet-not-showing-fixes covers the usual causes. Everything starts at codexpetgenerator.com.