It was late in 2023 when I first used ChatGPT, and I remember watching a tool generate content in front of me.

AI was already occupying most of my LinkedIn feed, including a good deal of fatalistic commentary about what was coming. I wanted to test it properly rather than argue about it. Work was not the place to start, so I stayed away from design, at least for a while.

The pattern that emerged over the next two and a half years was consistent. Each project began as curiosity about something I lacked the skills to build, and the value came from the process leading to the output.

The things I am building now derive from what I learnt along the way. This post is an introduction to what comes next.

Starting simple

My first experiment took an evening. Taking a small interest of mine, I set up a simple gardening blog on WordPress. Gardening had nothing to do with my work.

Within twelve days it was drawing an average of five visits a day, organically, with no promotion and no further effort from me. Small numbers, but more than I expected from a site I had done nothing to push.

The revelation was the speed. An idea reached production in a few hours rather than sitting unbuilt for weeks. However, blogging itself was not the interesting part. I wanted to build the things I had always wanted to build but lacked the skills and the time to learn.

Starting from a question

Fast-forward to April 2025, after trialing paid subscriptions to Gemini, Claude and ChatGPT, I began my first long-term project from a single question. What would it take to build a trading strategy I could run against the asset classes and products I already knew?

The workflow was straightforward. I set up a Claude project and grounded it in trading strategy knowledge and theory, operational costs and trading instructions, then used it to write strategies in ProRealTime’s PRC language that I could run in automation on the platform.

Over eight months of development and testing I released and trialled more than eighty strategies, predominantly on demo accounts. Trading performance across that period returned 18% profit but the return was not the only thing I was keen on assessing.

The process taught me more than the outcome did.

I learned to embed skills that improved my prompting and project setups. I learned to ground a prompt properly using markdown files. I tested a range of AI tools until I settled on a tech stack that gave me a faster and more reliable workflow.

New year, new projects

Towards the end of last year and into 2026, the approach changed. I was using AI to give my ideas substance rather than to generate content, and design became central to the work.

I was elaborating, adding depth, I was thinking in operational terms and iterating on those aspects at speed.

Given how much closer Claude and Figma had become, I started from design and thought about the product as a whole, going beyond the UI. For the first time I was setting up and connecting a database, an authentication layer and API connections, and assessing the security of each product before releasing it.

The projects below trace my journey as a maker. They are rooted in wanting to shape and create experiences, and in driving the changes that let me express my ideas.

SpellingGems.com

I am the father of two beautiful, clever and independent girls. Sofia, one of my two daughters, has recently been diagnosed with dyslexia. It is not great seeing your child struggle, expressing frustration when reading and going through that feeling of small failure. She has been trying very hard, so I felt I could help and do the same.

SpellingGems.com is my attempt to support my child with exercises that suit her learning journey, and to make that struggle less of a burden. It is my approach to making the experience more playful, making her part of the solution instead of the problem.

It is co-creation through ideation, design and testing. It is partly marketing and sales too, as I later found out she has been promoting the app with her school friends. Palydates have made this into user testing and research. It was chaotic at times, but I can’t wait to share some of that chaos here.

AlgoMaster

AlgoMaster is my attempt to evolve my interest in trading. There is no intention of becoming a new trading sensation. My only desire is to build an automated algo interface from scratch, starting simple with an MVP that included back-testing, then building complexity over time.

It is a system where the execution and monitoring layers work together at different speeds. It is still a work in progress, but I love it.

AntonioTrotta.me

Like millions of people out there, I have been fiddling with WordPress for years. Since my first attempt to use it in 2009 I have always played with the idea of creating my own template, but that desire never went beyond basic changes to pre-existing PHP code.

As I was contemplating the need to update my personal portfolio site, I asked myself how I could use AI to do something more elaborate. Could I build my own plugin? Could I define bespoke content blocks?

It is possibly the least fancy of all of them, but it is the one where I was able to trial interesting ideas with a pre-existing tool I had to learn and adapt to.

There are more, but these are the projects I want to share. All three have broken in ways I did not predict, and those are the parts worth writing about.