HumbleBee started at an airport

Some software ideas begin in a roadmap meeting. HumbleBee began after a cancelled flight.

I had been in San Francisco for the GitHub conference. Everything there is about AI, even the city was plastered with ads for AI. It was an eye-opener to see how AI was used among my US colleagues, and I promised myself I would seriously look into agentic coding.

I had a few great days and looked forward to going home. Then my flight home was cancelled.

With roughly eighteen hours at the airport and no place to sleep, I had all the time to think about the past few days. I met many great people, among them Carlos, the founder of GoReleaser. He was so passionate about his project that he made me think about using Go myself.

Then I had the idea: could I learn Go faster with AI? Could I also use my knowledge of time tracking to make something useful?

That is where HumbleBee started.

Go

Go had never been a strong interest of mine, but meeting so many people in San Francisco using it, I became curious.

They told me the code is readable and deployment is simple. It is also fast.

My experiment was supposed to become something like pair programming with an experienced developer. I didn’t want to “vibe code”, without understanding what happened. So everything I didn’t understand, the agent had to explain or point me to the right docs.

I had to slow the machine down, so it was not just throwing out code like… like a machine.

Why time tracking?

Time tracking is familiar territory for me.

I built Time & Bill back in 2011. I closed it in 2023, but now it’s back in service. It has support for teams, extended features like mileage logs and so on. However, when I reopened my doors I had to enable subscriptions.

HumbleBee should be a tool for people who care about privacy and open source. My first idea was to only make this tool for developers and provide only a CLI, but then I thought about the many emails from Time & Bill customers who asked me for a solution that didn’t require subscriptions. Now HumbleBee has a standalone GUI as well.

HumbleBee stores data in a local SQLite database. It can import Time & Bill exports. It has reports, database switching, and a workflow designed for developers, freelancers, consultants, and solo users.

What AI helped with

My learning loop using AI as a pair programmer and senior developer at my side was much faster than ever before. In the old days, I would have read a book and tried to apply the concepts in the order the author suggested. As a senior developer myself, I could already identify parts of the code and how it probably worked. Now I could go directly into the parts that didn’t make sense to me.

I also used it to give me recommendations on best practices. To my surprise, after asking for this, I found out AI would not always follow best practices itself.

What AI did not replace

The hardest part was still deciding what the product should be.

I could move forward quickly, since I already knew what went wrong with Time & Bill and what should be better with HumbleBee. Still, standalone apps are quite new territory for me, so there were some challenges for me there as well. Decisions are nothing AI could take away from me.

AI also did not replace reading code. In fact, I have never read as much source code as I did in the past month. All my previous knowledge about concepts and systems was still valuable when judging quality and identifying potential issues.

I also found that AI made me faster at learning and building products, but it’s also very tiring. If you don’t take care, the changes are huge and hard to review. Tiny commits, TDD and everything we have learned in the old days are still relevant.

From airport experiment to released software

What started as an airport experiment became a real release.

HumbleBee now can be installed using Brew or Scoop. You can also download the GUI and run it as an app.

If you are interested in the source code, or even want to contribute, I invite you to take a look at HumbleBee on GitHub.