Nobody Wanted to Write Release Note. So I Made a Git-Log Searching Tool Using AI Agent at Work.

Rondo — Mon, 13 Apr 2026 14:00:10 +0000

TL;DR

Inefficient workflow
QA had trouble with tracking issues(They didn't know what changed)
Developers didn't update release note for QA. We were busy dealing with our own tasks. (I know, it was just an excuse.)
Sales kept asking the same question again and again
So I made a git log searching tool(webpage).

Every firmware release triggered the same cycle:
“What changed?”
“Which features are included?”
“Does this affect our product? At which OEM?”

The reason why I couldn't do this manually

I was assigned a task to deal with the problem. But:

Too short deadline
Other ongoing tasks

I had to finish the task as soon as possible.

I don't usually use AI agents at work to deal with my tasks, unless the tasks that meet three personal criteria:
1. The task must be clear and pinpointed.

2. The task must be irrelevant to our base code. (I don't want AI agents mess up with the huge code base.)

3. The task must be what I can do on my own(So that I can fix problem later on.)
Fortunately, the task I was given met all the criteria.
So I thought it was the best time to try an AI agent at work.

The Tool to Make

A webpage that allows other teams to browse and search git logs(Of course, they're pre-filtered.)

Since I know a bit about Django, I decided to use that.
The setup process looks like this:

Clone repository on the server where the tool is running.
Save the path of the repository using django-admin.
Get git logs based on the saved path.

How I Built It

I decided to use Gemini CLI with iterative loop:

Ask Gemini-CLI to implement an exact feature.
Accept the output.
Run the code and check issues.
If any issues found, then let Gemini-CLI know the issues and let it fix the issues.
If Gemini-CLI keeps failing in finding the issues, then I fix them.

Example prompt:

Add a dropbox to limit shown logs on each page. The range is from 10 to 100 and gap is 10. The dropbox must be located next to filter box.

I also asked questions below multiple times during the process:

'Is the project really going well?'
'Are there any other ways to make it better?'

These kinds of meta-questions helped the agent stay on the right track.

One little trick that (might have) helped a lot:
I asked the AI to maintain a task log file to track progress and check it after every single task. Already know this trick? Sorry.

The Result

The final tool:

Select repository and branch
Filter by target product
Search commit messages (version, tags, etc.)
Pagination

Built with:

Django (and Django template)
SQLite
Local repositories + git log

Effects:

QA and Sales stopped asking repetitive questions.
Satisfied boss
Happy developers

It's not perfect.
Repositories have to be updated manually (git pull), but believe it or not, it's intended(sometimes you need to intentionally hide information.)

What I Learned

Using an AI agent didn’t remove the need to keep thinking.
Actually, I think it only made one skill more important:

The ability to define problems clearly and break them into small steps.

AI can generate code fast, of course.
But it won't decide what should be built and why.

Closing

This was a small tool, but a meaningful experiment.
I don’t think AI will replace developers anytime soon.
But that's only for those who won't stop thinking, I guess.

I'm a junior developer based in Korea, sharing what I learn and build along the way.
I also wrote a more detailed version on my blog: https://dev-rondo.com/posts/making-a-tool-with-agent

DEV Community: Rondo