My AI Agent - Introduction
This is something I’ve been wanting to do for a long time, but only now I managed to push myself into it.
I think the experience of working with AI agents has given me a lot of confidence on the quality of the outputs, and allowed me to enjoy thinking more about the product, strategy and code direction rather than actual implementation. I think this is the best moment in history to start this kind of experiment. We have all the tools and no excuses.
Specification Driven Development
I wanted to dive deeper into Specification Driven Development. I came across SPDD (which is similar to SpDD, I hope) through this post on Martin Fowler’s blog, and ever since I’ve fallen in love with this method of work — diving deep into specification before actual implementation.
It’s been enjoyable not just because its easier than coding youreself. It’s genuinely fun to think about product, and, most important, it’s rewarding. Thinking about the product is genuinely enjoyable, and most important: it feels rewarding. You get the speed of vibe coding, but instead of hoping the AI to go to the same direction you were thinking, you keep the accuracy that comes from actually doing things yourself. Because, in fact, you have detailed exactly what you wanted, and beyond, since the process forces you to think about gaps and edge cases long before implementation starts.
Why document this
I watched this video on the “anti vibe coding” strategy, and I thought that documenting my journey could both help me be more conscious of my decisions and my process, but also inspire others that it’s not only possible, but easy and fun. If it truly is, well, we will find out!
My constraints for this project
Useful for me, right now. My workflow will be to do something useful for myself and myself only. I refrain from thinking about scalability and multiple users. If in the future some people find it useful, they can adapt it for themselves — but that’s not what I’m optimizing for.
One feature at a time. Always something that is useful right now for me. No speculative features.
Not open source yet. I’m not sure about the privacy and security issues and I don’t want to think about that now. I’m not sharing my source code just yet.
Real deployment. I’ve never actually deployed a personal app by myself, and never used the express tools like the fallen Heroku and friends. I wanted to test this kind of little friction, high speed tooling.
The project
Well, shocker — it’s an AI agent.
I’m tired of having 3000 tools and integrating all by myself, or having to choose what to do where, what data to lose or manually integrate. I want it all centered.
I want a proactive agent, fast and flexible, able to help me execute my tasks and be aware of my agenda, goals, daily habits, projects, tasks, personal knowledge, and financial management.
It’s built around my needs, integrating all the tools that I have.
What’s next
This is the project introduction. The next post is about the first week of development.