AI made it easy
The time between imagining something (software in our case) and using a working version has been reduced by AI. It's not about creating a production-ready application in 10 minutes, that's still not possible, in my opinion. It's more about creating prototypes quickly.
I recently changed my way of developing from bottom-up to top-down: I have an idea, I ask the LLM to implement something (giving it a strong guideline) and I test whether the functionality I created is really worth it. If so, I go deeper into the implementation details. Before LLMs, most of the work was creating building blocks and connecting them.
Kaizen!
Every day is an opportunity to learn something. Every point of friction is an opportunity for improvement. I always struggled with how to efficiently manage my time. Now that I am a freelancer, I have to manage multiple small projects. Some customers use Jira, some others use different tools. I felt the need to use something of my own, something not shared with the customer, for example to handle personal tasks (like this blog).
That is where Ambrogio came in. The name is commonly used to indicate a butler, at least in Italy, where an ad in the 1980s popularized this name. I started Ambrogio as a simple bot that can read a large personal plain-text file I use to organise my life and store information (take a look at LLMs and one big text file).
I tried it for a couple of days and honestly, I found it was not so useful. I kept the functionality, and from time to time I still use it to look for specific information.
Then I moved to something I already knew was useful: a pomodoro timer. I've been a fan of Francesco Cirillo's work for many years, but I always struggled with all the apps I have used: some were too chaotic, some too minimalistic. Why not implement something shaped around me now that I have these superpowers?
Pomodoro + task manager
I became a fan of the command line. Using the keyboard a lot and the mouse less feels natural to me. I even miss smartphones with physical keyboards. I don't like tools with too many visual effects that can distract me. So, with my new friend Claude, I implemented a small CLI with a pomodoro timer inside Ambrogio.
The natural continuation was to implement a task manager. I have different customers, so it made sense to add the concept of projects. And if you have tasks, sometimes you have to take notes, for example about what you have done and the difficulties you found.
With my super-productive friend everything was quite smooth. The interesting part was that I didn't have to prioritise the design of the commands. I directly implemented them, used Ambrogio with the new features for a couple of hours and, if a feature was good, I refactored the code to fit my standards.
About the structure
My last experience with plain text files was great, so I continue using them for keeping the tasks and the pomodoros. The interesting part is that I can simply open the text file if I want to fix a typo, or in general change something. I don't need complex features implemented. Here is an example:
## Personal
- [ ] write a post about Ambrogio
- 📝 Ambrogio is the best tool I have ever used
- [x] pay taxes
- 🍅 2026-02-19 15:09
## Project 1
- [ ] Awesome feature
- 🍅 2026-02-19 15:19
- 🍅 2026-02-19 16:03
## Project 2
- [ ] Not so awesome feature
- 🍅 2026-02-21 15:59
- 🍅 2026-02-21 16:28
- [x] refactor the solution
Having fun with it
Building my own tool means I can do whatever I want, for example, why not make my office light blink at the end of a pomodoro? I built a hook I can use to trigger a behaviour and created a script to change my Philips Hue bulb color. That was it. It was fun and my daughters were surprised to see the light automagically change color without pressing a button.
I did the same with Claude, precisely when it finishes a task. That was also a lot of fun.
Is it because of LLMs?
In the end, working on Ambrogio has been interesting for two reasons: I can build a tool I use every day, and I can improve it quickly.
Before, I was not doing this because of all the setup work: boilerplate, integrations, and constant context switching. Most of the core logic is simple, but the setup can drain your time and energy.
This is where LLMs make a big difference for me: they reduce that friction, help me read docs faster, and connect concepts across tools. Modern stacks are now very complex, and this is the key point for me. So the real question is not if we should use LLMs, but how to reduce complexity so one person can still manage an entire system.
Thank you for taking the time to read this post. If you want to take a look at Ambrogio, here is the GitHub repository.