<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Blog – Rodrigo ES Porto</title><link>https://rodrigoesporto.com/en/blog/</link><description>Posts about my journey building with AI: Hermes, Grok, Gemini in Antigravity, Cursor and Hugo. Mistakes, costs and what I&amp;#39;m learning.</description><language>en</language><lastBuildDate>Sat, 03 Oct 2026 09:00:00 -0300</lastBuildDate><atom:link href="https://rodrigoesporto.com/en/blog/index.xml" rel="self" type="application/rss+xml"/><item><title>The beginning of a journey</title><link>https://rodrigoesporto.com/en/blog/the-beginning-of-a-journey/</link><pubDate>Sat, 03 Oct 2026 09:00:00 -0300</pubDate><guid>https://rodrigoesporto.com/en/blog/the-beginning-of-a-journey/</guid><description>
&lt;p&gt;Not long ago AI stopped being a distant headline and became a desk tool. I’m not “starting from zero.” I have a degree in Information Systems. I never coded as a job. I worked in sales and commercial representation, then in quality at an American multinational, with field procedures, new-product training and audits. Later I studied financial markets in search of asset diversification, a dividend portfolio and cryptocurrencies. That is not investment advice. Today I work as a ride-hailing driver and I build AI agents for my company. I’m coming back to tech, coding with AI’s help, after a career in sales, quality and process.&lt;/p&gt;
&lt;p&gt;I decided to write this in the open, for free. I don’t want to forget what I’m doing, and I want someone who isn’t technical to follow without translating jargon. On this site the reader will find simple guides, behind-the-scenes notes on what I’m putting together, and also the mistakes and the costs, without makeup.&lt;/p&gt;
&lt;p&gt;Why would someone who came from sales, quality and process have anything to teach people who aren’t technical? Because that career trained exactly what the tool does not do on its own: write the steps down, train new people without leaving them lost, and explain in plain language what still feels complicated.&lt;/p&gt;
&lt;p&gt;I want to practice four ways of creating with AI. They are not magic levels. They are four different doors, each with a daily use and a caution.&lt;/p&gt;
&lt;p&gt;The first way is asking. In practice, it is talking to the tool the way you ask a question at a counter. I use Grok in the app and inside X. It is for looking something up, drafting an idea, asking for an example, or unsticking a sentence. Day to day, that replaces part of the endless search and the loose notes. The caution is simple: an answer can sound sure and still be wrong. I still decide what counts, what to check, and what to throw away.&lt;/p&gt;
&lt;p&gt;The second way is handing work over. Here I don’t only ask for a paragraph; I ask for a task. With Grok Bot I give a goal and an agent tries the steps. One of them is called Darth. Day to day this is for repeated work: organizing, following up, drafting, gathering what I already know I have to do again. Handing over is not walking away. It’s asking for execution and then checking. The caution is not treating the result as finished just because it arrived formatted. If I don’t review it, I only swapped one extra job for another.&lt;/p&gt;
&lt;p&gt;The third way is building. In Grok Build I describe what I want and we try to get to an app. It is for turning a small idea into something you can open and use, without starting from code. Day to day that helps when the need is concrete: a flow, a screen, a way to store or show information. The caution is how clear the idea is. If I’m vague, the result is too. I need to say what goes in, what comes out, and who it is for, or the tool fills the gaps its own way, not mine.&lt;/p&gt;
&lt;p&gt;The fourth way is programming. That’s Cursor, real code, and what’s already live: this site in Hugo, published through Cloudflare and built with Gemini and Claude, inside Google Antigravity. For the company’s daily work I use Hermes Agent on the laptop and in the cloud, with task agents and a newsroom that helps me write. Programming, in this sense, is when the conversation becomes a file, a page, an agent or a site that still exists after I close the chat. The caution is the same as any craft: understand what was done well enough to fix it when it breaks, and not pretend “AI handled it” if I don’t know what is running.&lt;/p&gt;
&lt;p&gt;None of this replaces judgment. It shortens the distance between “I know what I want” and “it’s live.”&lt;/p&gt;
&lt;p&gt;If all of this sounds basic to people who already live in software, good. This piece isn’t for them. It’s for people who drive, sell, run a process, or only have an hour at night and still want a decent tool without feeling stupid.&lt;/p&gt;
&lt;p&gt;I’m curious. I look things up and I study new tools. I want to help. I teach for free what I’m doing so people who aren’t technical can use what makes daily life easier. No course for sale.&lt;/p&gt;
&lt;p&gt;Follow along on X, &lt;a href="https://x.com/RodrigoesPorto" target="_blank" rel="noopener"&gt;@RodrigoesPorto&lt;/a&gt;, and here at &lt;a href="https://rodrigoesporto.com" target="_blank" rel="noopener"&gt;rodrigoesporto.com&lt;/a&gt;. I’ll post here as I go, with the wins, the mistakes and the costs.&lt;/p&gt;</description></item><item><title>Why I started and the four ways I create with AI</title><link>https://rodrigoesporto.com/en/blog/why-i-started-and-the-four-ways-i-create-with-ai/</link><pubDate>Sat, 03 Oct 2026 00:00:00 +0000</pubDate><guid>https://rodrigoesporto.com/en/blog/why-i-started-and-the-four-ways-i-create-with-ai/</guid><description>
&lt;p&gt;My name is Rodrigo ES Porto. I live in Vitória, in Espírito Santo, Brazil. This is the first piece in the Jornada series: a public diary, one piece a day, of what I am learning with artificial intelligence.&lt;/p&gt;
&lt;p&gt;The question here is simple. Why I started, and which four ways I use to create with AI. A lot of people only know the chat that answers a question. I started there too. Later I saw you can go further without pretending you are a professional programmer.&lt;/p&gt;
&lt;p&gt;I teach for free what I am learning. I do not sell a course. If I cannot explain it simply, I do not understand it yet.&lt;/p&gt;
&lt;h2 data-hextra-search-id="what-i-wanted-to-fix"&gt;What I wanted to fix&lt;span class="hx:absolute hx:-mt-20" id="what-i-wanted-to-fix"&gt;&lt;/span&gt;
&lt;a href="#what-i-wanted-to-fix" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;I have a degree in Information Systems. I worked in sales and commercial representation. Later I worked in quality, training, and auditing at an American multinational. I never programmed for a living. Today I drive for a ride-hailing app, Monday to Saturday, from morning until late afternoon. At night and on Sunday I work with AI agents.&lt;/p&gt;
&lt;p&gt;What I wanted to fix fit in a few lines: get the routine in order, look after the profile on X, and stop letting ideas die. And come back to technology through study and AI.&lt;/p&gt;
&lt;p&gt;In early September 2026 I opened Grok Bot, Grok&amp;rsquo;s app for agents, and I made a main assistant. It works like a chief of staff: it takes what I ask and coordinates the rest. That was the starting point. The rest of the series tells what came after, one piece at a time.&lt;/p&gt;
&lt;h2 data-hextra-search-id="quick-summary"&gt;Quick summary&lt;span class="hx:absolute hx:-mt-20" id="quick-summary"&gt;&lt;/span&gt;
&lt;a href="#quick-summary" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;I came back to technology with study and AI, driving by day and studying at night.&lt;/li&gt;
&lt;li&gt;AI, in this text, is a program that suggests the next word, the next step, or a piece of code from what you ask. It is not a person.&lt;/li&gt;
&lt;li&gt;The four ways I use are ask, delegate, build, and program.&lt;/li&gt;
&lt;li&gt;None of these tools replace judgment. They shorten the path from &amp;ldquo;I know what I want&amp;rdquo; to &amp;ldquo;it is live.&amp;rdquo;&lt;/li&gt;
&lt;li&gt;The first practical lesson: write what the agent may and may not do, and ask for a short answer, to save time and usage.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 data-hextra-search-id="first-what-artificial-intelligence-is"&gt;First: what artificial intelligence is&lt;span class="hx:absolute hx:-mt-20" id="first-what-artificial-intelligence-is"&gt;&lt;/span&gt;
&lt;a href="#first-what-artificial-intelligence-is" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;When I say AI, I mean the program that tries to continue what you wrote. The picture that works for me is the suggestion bar on a phone keyboard. You type two words and it guesses the third.&lt;/p&gt;
&lt;p&gt;Inside, the mechanism is similar, only much larger. The model has seen a huge amount of text. Faced with your request, it calculates what is most likely to come next: a word, a sentence, sometimes a whole plan.&lt;/p&gt;
&lt;p&gt;The comparison stops at one point. Phone suggestions do not organize your week and they do not build a website. And AI does not &amp;ldquo;know&amp;rdquo; whether the output is true. It is wrong with a straight face. I am the one who checks.&lt;/p&gt;
&lt;h2 data-hextra-search-id="first-what-an-agent-is"&gt;First: what an agent is&lt;span class="hx:absolute hx:-mt-20" id="first-what-an-agent-is"&gt;&lt;/span&gt;
&lt;a href="#first-what-an-agent-is" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;An AI agent is an assistant with a written role. In a normal chat, you ask and you read the answer. With an agent, you give a goal and it tries to get there in steps.&lt;/p&gt;
&lt;p&gt;I think of a willing intern who follows the instruction to the letter. If the instruction is vague, it improvises. If the instruction is clear, it actually helps.&lt;/p&gt;
&lt;p&gt;The comparison breaks here: the intern gets tired, asks for a hand, and leaves at six. The agent keeps going until the limit you or the tool set. That is why I write rules. Without rules, it rambles. With rules, it fits in my evening.&lt;/p&gt;
&lt;p&gt;Grok is the assistant from xAI. I use it in the app and on X, which is the only social network I am on. Grok Bot is the app where those agents get a role and start receiving a goal, not only a question.&lt;/p&gt;
&lt;h2 data-hextra-search-id="the-four-ways-to-create-with-ai"&gt;The four ways to create with AI&lt;span class="hx:absolute hx:-mt-20" id="the-four-ways-to-create-with-ai"&gt;&lt;/span&gt;
&lt;a href="#the-four-ways-to-create-with-ai" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;I did not invent a theory. I named what I was already doing, so I would not mix the tool with the expectation.&lt;/p&gt;
&lt;h3 data-hextra-search-id="1-ask"&gt;1. Ask&lt;span class="hx:absolute hx:-mt-20" id="1-ask"&gt;&lt;/span&gt;
&lt;a href="#1-ask" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h3&gt;&lt;p&gt;Tool: Grok, in the app and on X.&lt;/p&gt;
&lt;p&gt;This is for a doubt, a search, or an idea that is stuck. It is the closest to a conversation. I write the question, read the answer, check it, and go on. It does not publish by itself, and it only becomes a file if I copy it.&lt;/p&gt;
&lt;p&gt;If the subject is serious, I do not stop at the chat. I open the source. Grok gets the idea moving. Checking is my job.&lt;/p&gt;
&lt;h3 data-hextra-search-id="2-delegate"&gt;2. Delegate&lt;span class="hx:absolute hx:-mt-20" id="2-delegate"&gt;&lt;/span&gt;
&lt;a href="#2-delegate" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h3&gt;&lt;p&gt;Tool: Grok Bot.&lt;/p&gt;
&lt;p&gt;Here I pass a goal and an agent tries to do the steps. Delegating is not walking away: it is asking and then checking.&lt;/p&gt;
&lt;p&gt;In early September, the goal was the routine, the profile on X, and the ideas. The main assistant took the request and coordinated the rest. I did not let go. If I do not read what came back, I did not delegate. I walked away from the work.&lt;/p&gt;
&lt;h3 data-hextra-search-id="3-build"&gt;3. Build&lt;span class="hx:absolute hx:-mt-20" id="3-build"&gt;&lt;/span&gt;
&lt;a href="#3-build" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h3&gt;&lt;p&gt;Tool: Grok Build.&lt;/p&gt;
&lt;p&gt;I describe what I want and try to get to an app. It is still a conversation, but the result wants to be something you use, not only a paragraph.&lt;/p&gt;
&lt;p&gt;You can picture it as ordering furniture by describing the room. In practice, the tool generates what it understood from the request. The picture fails if I do not know what I want: the thing comes out crooked and I lose the evening.&lt;/p&gt;
&lt;h3 data-hextra-search-id="4-program"&gt;4. Program&lt;span class="hx:absolute hx:-mt-20" id="4-program"&gt;&lt;/span&gt;
&lt;a href="#4-program" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h3&gt;&lt;p&gt;Tools: Cursor and Google Antigravity, two code editors with AI. Real code: a file, a site, or an agent that still exists when I close the chat.&lt;/p&gt;
&lt;p&gt;Here the conversation becomes something that stays. I stay in command. I never programmed for a living. AI writes fast. I decide what gets in. The rest of that story belongs to later pieces. Today what matters is the map.&lt;/p&gt;
&lt;h2 data-hextra-search-id="how-i-pick-the-way"&gt;How I pick the way&lt;span class="hx:absolute hx:-mt-20" id="how-i-pick-the-way"&gt;&lt;/span&gt;
&lt;a href="#how-i-pick-the-way" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;My filter, today, is this:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;If I have a doubt, I ask Grok.&lt;/li&gt;
&lt;li&gt;If I have a goal with several steps, I delegate in Grok Bot and I check what came back.&lt;/li&gt;
&lt;li&gt;If I want an app from a description, I try Grok Build.&lt;/li&gt;
&lt;li&gt;If it needs to keep existing as a file, a site, or an agent, I go to the editor, in Cursor or Antigravity, and I keep the command.&lt;/li&gt;
&lt;li&gt;In any of the four, if I cannot check the result, I do not go ahead.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;When it works, I read it at night, check it, and use it. When it does not, the answer comes long and the evening is gone.&lt;/p&gt;
&lt;h2 data-hextra-search-id="what-i-used"&gt;What I used&lt;span class="hx:absolute hx:-mt-20" id="what-i-used"&gt;&lt;/span&gt;
&lt;a href="#what-i-used" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;At the start, almost everything went through Grok. The chat in the app and on X to ask. Grok Bot to delegate. Grok Build to build. Cursor and Antigravity come in when the subject turns into a file.&lt;/p&gt;
&lt;p&gt;I use these tools in my own work. I do not speak for any company. When a company promises something, I treat it as their promise, not as a fact of mine.&lt;/p&gt;
&lt;h2 data-hextra-search-id="what-went-wrong"&gt;What went wrong&lt;span class="hx:absolute hx:-mt-20" id="what-went-wrong"&gt;&lt;/span&gt;
&lt;a href="#what-went-wrong" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;It was not a screen error. It was a vague request. Without saying what the agent may and may not do, it improvised. Without asking for a short answer, the text ballooned. I drive all day. I get home with no patience left for a machine novel.&lt;/p&gt;
&lt;h2 data-hextra-search-id="what-i-learned"&gt;What I learned&lt;span class="hx:absolute hx:-mt-20" id="what-i-learned"&gt;&lt;/span&gt;
&lt;a href="#what-i-learned" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;The earliest lesson was not technical. It was writing. I had to write clear rules for the agent: what it may and may not do. And I had to ask for short, direct answers, to save time and usage.&lt;/p&gt;
&lt;p&gt;The other lesson came attached: delegating without checking is walking away. I do not want an assistant that publishes my life. I want a chief of staff who prepares, and a human who signs.&lt;/p&gt;
&lt;h2 data-hextra-search-id="what-i-would-do-differently"&gt;What I would do differently&lt;span class="hx:absolute hx:-mt-20" id="what-i-would-do-differently"&gt;&lt;/span&gt;
&lt;a href="#what-i-would-do-differently" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;Write the rules on day one. Ask for short from the start. Do not wait for the evening to turn into a tangle before seeing that the problem was the request, not the tool.&lt;/p&gt;
&lt;h2 data-hextra-search-id="costs-and-limits"&gt;Costs and limits&lt;span class="hx:absolute hx:-mt-20" id="costs-and-limits"&gt;&lt;/span&gt;
&lt;a href="#costs-and-limits" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;In this first piece I will not pin a tool price that I do not have on the table here. The cost I can pin is time: nights and Sunday. The &amp;ldquo;usage&amp;rdquo; I mention is how much the tool spends on each request. A short answer costs less of my time and less usage.&lt;/p&gt;
&lt;p&gt;The limit I can pin is judgment. The tool shortens the path. I publish. I sign what goes out under my name.&lt;/p&gt;
&lt;p&gt;Another limit: I teach what I am learning. If I am still in the middle, I say I am still in the middle. Mistakes and costs, with no dressing up.&lt;/p&gt;
&lt;h2 data-hextra-search-id="who-this-is-for"&gt;Who this is for&lt;span class="hx:absolute hx:-mt-20" id="who-this-is-for"&gt;&lt;/span&gt;
&lt;a href="#who-this-is-for" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;This is for someone who is not a specialist, has little time, and will check what the machine returns.&lt;/p&gt;
&lt;p&gt;It is not for someone who wants a button that fixes their life, or for someone waiting for a method for sale. It is also not a veteran programmer tutorial yet.&lt;/p&gt;
&lt;h2 data-hextra-search-id="closing"&gt;Closing&lt;span class="hx:absolute hx:-mt-20" id="closing"&gt;&lt;/span&gt;
&lt;a href="#closing" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h2&gt;&lt;p&gt;I started because the routine needed help and because I wanted to come back to technology without wearing a badge I do not have. The four ways (ask, delegate, build, program) are the map for this series. The first real step was an assistant in Grok Bot, in early September.&lt;/p&gt;
&lt;p&gt;In the next piece I tell that first assistant properly: how to delegate without walking away.&lt;/p&gt;
&lt;p&gt;I am on X: &lt;a href="https://x.com/RodrigoesPorto" target="_blank" rel="noopener"&gt;@RodrigoesPorto&lt;/a&gt;.&lt;/p&gt;
&lt;h2 data-hextra-search-id="sources-and-links"&gt;Sources and links&lt;span class="hx:absolute hx:-mt-20" id="sources-and-links"&gt;&lt;/span&gt;
&lt;a href="#sources-and-links" class="subheading-anchor" aria-label="Permalink for this section"&gt;&lt;/a&gt;&lt;/h2&gt;&lt;ul&gt;
&lt;li&gt;Grok and xAI: &lt;a href="https://x.ai/" target="_blank" rel="noopener"&gt;https://x.ai/&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;Grok Bot (company page): &lt;a href="https://x.ai/news/introducing-grok-bot" target="_blank" rel="noopener"&gt;https://x.ai/news/introducing-grok-bot&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;Grok Build (company page): &lt;a href="https://x.ai/build" target="_blank" rel="noopener"&gt;https://x.ai/build&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;Cursor: &lt;a href="https://cursor.com/" target="_blank" rel="noopener"&gt;https://cursor.com/&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;Google Antigravity: &lt;a href="https://antigravity.google/" target="_blank" rel="noopener"&gt;https://antigravity.google/&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;My profile on X: &lt;a href="https://x.com/RodrigoesPorto" target="_blank" rel="noopener"&gt;https://x.com/RodrigoesPorto&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;</description></item></channel></rss>