Why I Started Building AI Agents

I remember the exact moment I got hooked.

I was watching a demo where a language model was given access to a search tool and a calculator. The prompt was something mundane — “what’s the GDP of Germany divided by its population?” — and instead of hallucinating an answer, the model stopped, reached for the search tool, fetched live data, passed the numbers to the calculator, and returned a correct, cited answer.

It sounds small. But I couldn’t stop thinking about it.

The Shift from Chatbot to Agent

A chatbot responds. An agent acts. That distinction sounds philosophical until you see the difference in practice. A chatbot asked to research a topic will confidently make things up. An agent with a web-search tool will actually go look.

The tools don’t have to be exotic. File I/O, an email API, a database query — once a model can reach outside its context window and affect the world, the category of problems it can solve changes entirely.

Why a Hobby Project?

I build these at nights and weekends. No product requirements, no deadlines. That freedom means I can follow curiosity wherever it leads — which turns out to be the best way to learn how agents actually fail.

And they fail a lot. In ways that are fascinating.

Watch this space.