When I took the plunge and acquired the Google AI Pro (5 TB) plan, I confess I felt that euphoria of wanting to take on the world. In my head, the idea was straightforward: “With this, now I’m truly going to develop and build things non-stop, super fast and almost on autopilot”.
I thought artificial intelligence was going to solve almost everything just by typing a couple of lines.
But reality did not take long to bring me back to earth.
The illusion that AI “guesses” what you have in mind
At first, I fell into the typical mistake: I threw quick text prompts expecting the agent in Antigravity or the chat to guess exactly the logic I already had structured in my head. I wanted it to understand the business context, database details, and how I like to organize code purely by intuition.
Obviously, it didn’t work that way.
It handed me things that seemed fine at first glance, but when I stepped in to review, they didn’t fit the way we build things in the project. I had to correct, ask again, and in the end I felt like I was wasting time.
That’s when I understood something key: AI is not a fortune teller that reads minds. It is a powerful tool, but it needs you to give it order and direction. If you aren’t clear on the flow, the machine will be even less so.
The advantage of thinking like a programmer
When I realized that, I changed my approach. At the end of the day, I am a programmer and my mindset is structured by nature.
We developers are used to following a logical flow:
- Defining what goes in and what should come out.
- Anticipating errors before they happen.
- Breaking a big problem down into small parts that can be solved one by one.
Instead of asking it to make a complete module all at once, I began guiding it step by step. And that’s where the experience took a complete turn: adapting to that workflow was very natural because it is the very nature of programming.
The real change: Working with my own Skills
Constantly explaining to the agent over and over how I like to structure things in every session was tiresome. That’s why I began building and refining my own custom Skills inside Antigravity:
- Clear rules so components remain clean, organized, and without shortcuts that cause problems later.
- Firm standards for database procedures with safe transaction handling.
- Architecture guidelines so new code always respects the structure of what is already working.
Nowadays I no longer have to repeat context from scratch. The agent reads my skills, knows what standards to follow, and generates code aligned with my way of working. That changed the game completely.
Lightening the load: Processes that on paper seemed impossible
One of the greatest reliefs I have felt in my daily work is how it lowers mental load.
Sometimes I had requirements at work that I pictured in my head as extremely heavy. I would sit down to sketch them on paper and think: “Uff, assembling this whole workflow is going to take me days of mental exhaustion”.
But sitting down with the tool, bouncing the idea around, and structuring it alongside AI, I have realized that many processes you imagine as monsters can actually be solved in a much simpler and more elegant way. It removes that initial block and helps you see the most direct path.
Delegating the heavy lifting and thinking a step ahead
Nowadays I work at a much more agile pace because I delegate the heavy lifting to it:
- I define the logic and architecture: I know what the system needs to do and how it should behave.
- The agent does the repetitive work: It generates the base structure, forms, mappings, and the code that is time-consuming to write by hand.
- Quick review: Just by looking at the implementation, I know if it’s at 95% or almost 100%, and I only step in to make fine-tuned precision adjustments.
- Thinking about the next step: While the agent is processing or applying changes, my head is already free designing what comes next.
Deliveries are much faster and you work with less fatigue.
Where I stand today
Currently, my day-to-day relies on:
- The Google AI Pro (5 TB) plan as the foundation for the entire ecosystem.
- AI chat: To land raw ideas and validate approaches before touching code.
- Antigravity: As my daily workspace where the agent executes directly on files.
I am very aware that I still don’t use all the features the tool has; there are many advanced options I am still discovering. But for my everyday work, I clearly feel I am leveling up. When I have questions, I lean on online content and documentation, testing and adapting what works for me.
In the end, artificial intelligence didn’t come to program for me or replace what I know. It works for me because I know what I’m doing, I have clear fundamentals, and I know exactly what to ask for. It is a tremendous accelerator when you learn to stay in control.
