Techne

This one is about the time I moved from Lovable to Cursor.

I had all my projects inside Lovable, and everything was already on GitHub. The code from Manus, all the prototypes I was coming up with in Lovable. I had at least the basic understanding that the versions were there. So that part was done.

Then it got to the point where we had to decide between two apps: the CrossFit one or the mental health one. They were basically done. The design, the buttons, the colours, the flows. Everything was working.

But I didn’t have a login, and I didn’t have a database. I didn’t know how to do one.

This is for a developer

I knew I needed to learn it. But that went back to the block I had before: this is not something I can do. This is for a developer. I’m not a developer, I don’t understand databases, I’m not going to be able to do it.

I really remember opening Supabase, looking at it, and thinking: how am I going to do this?

I had some basic understanding of tables. I had done some very basic SQL. Most of the time it was me explaining what I wanted and telling a programmer or a data analyst to actually do it. Even in my PHP times. Even when I was doing SEO, I would sit by a developer, and the developer would implement it. I would just say what needed to be done.

So I had no real experience storing a database, or storing logins, or thinking about security. About what could happen. Much later, when I did the security review, I realised that the code I had guided the AI to write was not good.

One more move

It got to the point where I needed to publish. My co-founder in capivv said: just go for Cursor. It’s smarter than Lovable. It’s going to be a stepping stone. You’ll learn more, and it will be easier to understand how things work.

But moving to Cursor was making that move again. Manus to Lovable, then Lovable to Cursor. And the app was on me. He was building the tool, I was in charge of the apps, and that meant publishing them to the App Store.

I had done one publishing myself, manually, with ChatGPT as a guide. Copying, Ctrl C, Ctrl V, opening the terminal, putting in the commands. It was pretty scary the first time, but I got one version in through Xcode.

The terminal has always felt more familiar to me, because I had a Linux phase at the very beginning of the internet. That was a very specific moment in my life, but it still feels like a better place for me.

The database was different. I opened Supabase and I was completely frozen. I looked at the tables, I looked at the setup, and I didn’t even think of just asking ChatGPT for directions. I was frozen. And then the idea of moving to Cursor was another freeze, because moving platforms felt like such a hassle.

And then one day I just subscribed to Cursor. I usually do that when I’m very scared of something, or I don’t want to do it. I just do it, and then I take the hit.

A huge, huge list

The first thing I wanted was to put it to the test. He said it was smarter. So I took the code from viral hawk, the site I was developing for YouTube, and I said: check it out for me. What are the biggest mistakes, the biggest issues?

It came back with a huge, huge, huge list.

This was when Claude Code was starting to be a thing, and people kept telling me to move to it. I didn’t want to use Claude. When I started with ChatGPT I had checked all the AIs, and the first thing I asked Claude ate all my credits in two seconds. I had no sympathy for Claude.

But Cursor pre-selected Claude. So it was Claude that did the analysis.

So I wasn’t completely wrong. Yes, I made many mistakes. But the quality of the code was very bad, or at least Claude thought it was. It basically said I should develop it from scratch.

I was very, very sad. Two months and a half, building the scraper, the rules, everything. In my head it was a completed piece of work that needed small changes here and there. And suddenly it’s scrap. That’s fine. That’s part of the learning, I guess.

I ran the same thing on Interrupt and WODa Flow. Many issues again, some of them from Lovable. And then I ran out of tokens. The first time, in your face. Claude, again. No sympathy.

It kept changing the model

That is what I got really annoyed with in Cursor. It would go straight to a free model. You change to Claude or ChatGPT, and you run out of tokens. Sometimes it changed the model mid operation. You get stuck a lot.

And I needed to launch Interrupt. I needed it in TestFlight.

At that point my partner built an automation I still use today. It does every step of getting the app published. I had to do the building, the API keys, all the authorisations, to get the app into an automation on GitHub. Now I send a comment to my Claude and it publishes the app to TestFlight, and I can test it. Doing it manually was a big, big hassle. It is the best automation ever.

How lost I was

My friend said: if Cursor is not the one for you, maybe go to VS Code. My setup is VS Code, with Claude, and I work inside it.

One more platform change. So why not? But I was so not going with that. Cursor already felt like the place I could be.

And that was the moment I realised how huge the gap in my knowledge was. MCPs were starting to be a thing, and I sincerely didn’t know what an MCP was. I had to set one up inside Claude, and I didn’t know how. Again, the block: this needs a technical person.

I was so lost. Eventually I got the MCP working, and my friend explained what MCPs were. He was a developer with all the patience in the world, and we had a project together, so I could ask him how to move forward.

For the first time, I felt like: okay, now I’m becoming a little bit more literate. I also had to use the terminal inside Cursor, and that was quite interesting. Closer to my Linux times.

That’s when I thought: maybe this is not that hard. I’m just making this harder than it needs to be.

A hundred and fifty versions

I got some MCPs working, and I started testing things. And I did, I don’t know, a lot of versions of WODa Flow inside Cursor.

You get a certain number of minutes on GitHub for actions. I used them all. Every time I made a small change, I did a TestFlight version. Instead of looking at a feature, getting the feature right, having a plan, a roadmap, even designing by hand, which I do now because it is clearer for me and for the AI what I want to build.

I would change a text, and I would do a TestFlight version. That was crazy. WODa Flow has a hundred and fifty versions.

We hadn’t chosen which app to focus on, so I was improving both. When we decided on Interrupt, the mental health one, I started doing versions for Interrupt. Around fifty. Much less, but still a lot, because I was still in that mindset: every small change, a TestFlight version.

Another chapter

At some point Interrupt was pretty much done. There were not going to be a lot more changes. And the things I still needed to do, Cursor kept changing the model on me. I was pretty upset.

So I moved to VS Code. And the first thing I did was: let’s do a checklist of everything we need to launch on the App Store.

VS Code was the beginning of understanding AI in a bigger development sense. Seeing all the things I had done so far shown to me in a way that made sense as a project.

But that’s another chapter.

Techne, episode three. The Lovable months.

This one is about Lovable. Maybe a little more than a lot of people would expect, but mainly because it took me such a long time to realise I could move to other tools.

I think a lot of us that are not technical people think, all the time, that we are not going to be able to do it. That is the biggest fear. You look at your screen, something goes wrong, and even if you turn it off, something bad is going to happen. You won't know how to get away from it. You won't know how to restart. You are going to lose all your work, and you are going to get a virus. Especially if you are a millennial, if you watched all the hacker movies.

In your mind, anything a little more technical needs you to know how to code, or to understand databases and statistics and math and calculus. And you just go crazy thinking this is something you will need.

The reality is much less than we think.

I'm not saying you can get away with not knowing anything. You cannot. You will eventually have to learn things. But they come naturally. For me it became something I didn't even feel was happening to me.

All the ideas became possible

For the longest time I really did want to develop my own app. I worked a long, long time doing marketing for apps. Spending money on promotion, on distribution, on PR, buying downloads, buying everything. And I always found it interesting to know how things work from the ground up.

But I didn't want to spend the time learning how to code, nor did I have the time.

Then Lovable became possible, and all the ideas of all the apps I had ever had became possible too. And suddenly I was spending a lot of money on Lovable.

I was working with both platforms for a while, Manus and Lovable. By then I was one hundred percent sure Manus was lying to me and making me spend more money. I did it to myself. But I still think it's very possible. Allegedly.

So I started to just come up with app ideas. I had, I don't know, ten possible apps. I had the dashboards. I tried things out.

Where will things be stored?

Then it got to a point where I had to become a little more technical. Not because I was doing a coding course. I was just thinking: if I have a user, how will that happen?

How will they log in? After they log in, where will the information be? Where will things be stored?

All of these became things I needed to learn. But there was still the block. The block of thinking I needed to be very good at SQL. The block that I didn't study software development, I didn't know how things were, I didn't know how to do this.

At that point I was starting a conversation with a partner about a tool that would become a layer of revenue for app developers. That is capivv. It was very much just an idea we were discussing, and we needed an app to test it on. So: let's develop our own app. And I was the one who had to do it.

I had all these prototypes and logos. One of them, WODa Flow, a CrossFit app, I was one hundred percent sure would be the one. I had the screens, the whole branding part.

And then I came up with another idea. An app called Interrupt. Just the idea that you could quickly stop an urge. I thought: this is a very simple app. Simple to develop, simple to publish, and I have another project to promote it.

So I had two candidates, and a lot done for both. The screens, everything. What I didn't know how to do was the login, and where to store the information. The database. The backend parts.

And it was the time when a lot of social media was talking about Lovable projects having issues with databases. I was already scared, and I was blocked about the next step, which is actually having the app in the App Store.

So I asked my friend, the one I'm developing capivv with: what do I do?

He said: I think you need to change to a model you feel more safe using.

Because I had had the issue with Manus. And I was having the same difficulties with Lovable. I wanted to change something, and something else happened.

It was a checkbox

At that point I didn't understand that a lot of the time you just need good references. You need to be able to say what you want. Not giving the AI references and directions, it will be tricky.

One thing I find very interesting. When I was doing the Lovable course, we were in a group project, and we had checkboxes on one of the screens, and Lovable showed them in a way that didn't look like checkboxes. No matter what we did, we couldn't change it.

Then we talked to the teacher of the course, and he simply opened the code. He looked, and he saw it was a checkbox.

That is the big aha moment you start having when you use AI to develop your own projects. Eventually you look at the code and, organically, say: oh, something's not right here. You might not know what it is. But you know something is not working. Not because it's not done. Because it's not clearly shown to you.

This is just a platform that keeps my project

When I moved from Manus to Lovable, I had to migrate all my projects to GitHub. That was the beginning of using GitHub. Bear in mind, I had never used GitHub before.

I migrated all of them. Even all the other crazy projects I had inside Lovable. Even the SaaS one, which was a big, big surprise many months later, when I actually analysed the code. But that's another conversation.

I connected Lovable to GitHub, and I could see the projects. I made the accounts, I opened it, I looked at it, and I thought: this is just a platform that is going to keep my project. That's fine.

Fifty-five builds

The focus was: publish Interrupt. Much harder, because I had no idea how to work with the App Store, TestFlight, Xcode.

So I asked ChatGPT. I had the project inside ChatGPT, organised. I had realised I needed a new window for each feature. But I still hadn't realised I needed one major place the AI could visit. A major document. Your roadmap. Ideally your PRD. That is what stops the AI losing the context of the project.

That took me months to realise. And I didn't have anyone or anything to guide me.

I'm a person who needs to learn through projects. I'm not very good at learning things that take longer to apply. I should be. Our education is usually like that: you study something and later you apply it. It doesn't work like that for me.

I need to be inside the chaos. And then inside the chaos, I find my way out of it.

All of this was done with ChatGPT and the terminal open. I had used the terminal a long, long, long time ago. It was pretty much copying code from one platform to the other, making mistakes. And eventually I had the first version of Interrupt on TestFlight.

I didn't know how to do versioning. I didn't know how to do features. So every time I changed something, I published a new version from Xcode. WODa Flow went to around a hundred builds, maybe more. Interrupt went to fifty-five. The fifty-fifth build is the one that was published.

It was very funny, because I had no idea about anything. I was just doing things and breaking things and making it happen.

The issues on GitHub

There was a point where I was on GitHub and still thought I needed an Asana or some task planner. And then my partner said: look, you can use issues on GitHub.

Your roadmap gives you tasks. Your tasks become pull requests when you change your code. And you have the record, all of it managed through issues.

I was like: oh, interesting. But I still didn't understand how to use it, until I had to. On capivv I am the product person, not the developer. On Interrupt, I was the developer.

Around March I was setting up to move from Lovable to Cursor. The block was still in my head. I was still in a platform, not in a coding environment. I could see the code being done, but I was not ready to go there.

And my friend said: I think it's time. There is a lot of database stuff you are going to have to do, and I need to see what you are doing, so if there is a conflict with capivv I can step in.

So if I found an issue in capivv, in the connection with Interrupt, I had to report it as a GitHub issu…

Techne, episode two. The Manus months.

Manus came to me right at the beginning of it. I had some Chinese friends talking about it, saying it was great, it was different from ChatGPT. I was already prompting ChatGPT and trying things out, but nothing very clear yet. Just the Lovable project here and there. And again: no PRD, no path, no map, nothing.

I only found out I was not doing anything in depth much later, when I had a conversation with a very technical friend and asked him to check my PRD. Because I thought I had a PRD.

He said: why is your file so small?

I said: what do you mean, so small? In my conception that was a huge thing, full of details.

And he said: you don’t have references, you don’t have a roadmap, you don’t have a step by step, you don’t have anything that is actually a PRD.

I had no idea that was so basic.

The tool everyone wanted

Manus was the cool thing. It was difficult to get into. You needed an invitation, so not everyone had access. And the results were genuinely good. I had this huge research project it did for me and it did it really well.

The coding part I had no idea about, obviously. There was no Claude Code yet. Codex came much later. But it was the first time I could see something happening in front of my eyes, and it was different from Lovable. With Lovable you are prototyping, you are not really programming anything.

I started using it as a building tool around November.

What I was trying to build

My husband has a YouTube channel and was getting a lot of hits on shorts. He wanted research on what goes viral. So I needed a way to get him ideas and clips, which meant a ranking process, ranking videos by virality.

The idea was simple. Something that is just a radar. Press one thing, it looks up what is trending across platforms.

That is where I found out that getting information out of Instagram is much harder than YouTube, which you can scrape via API.

I had never used an API before. Maybe I had generated a key here and there, in meetings, with partners. But me, myself, generating an API key. That was the first time.

And I did it. No trauma. It was pretty easy.

TikTok and Instagram were not straightforward, and I thought the cost of solving that was too high, so I gave up and just did YouTube. Today I know how to go about it. At that point I had no idea.

Bear in mind: today you can build this in a day. In November 2025, for me, that meant many, many days. Months, actually.

The window that was being eaten

As most of you know, the context window matters. The amount of information you put into a conversation degrades it after a while.

I was having enormous conversations in one Manus window. And I had no idea what a context window meant.

So there I was, developing an entire tool inside a window that was being eaten. Eight hours a day, completely fixated, certain this was going to be my first SaaS.

I was doing all of it thinking it would be easy. When it got to the point where I needed some infrastructure, I would learn it as it became necessary. Users inside my platform only came up around the end of December. I thought: this is more or less ready for users. And then, after building everything, I wanted to add tenants.

It was just a mess.

The security check

I started going to ChatGPT because I thought Manus was not getting me what I wanted. At some point I said: I want to run some security checks on viral hawk.

It came back with so many issues. So many issues.

Open to SQL injections. Leaking information. All of the things you hear about vibe coding if you follow any of those people online, which I did, and I do.

And I was genuinely surprised. Because I was certain Manus was paying attention to this. You are vibe coding with no experience, so you have this idea that the tool does the basic things for you.

But you never asked for it.

I never asked Manus to build me the best tool for YouTube clippers that is also safe and also has a login area. I just kind of wanted the AI to read my mind. And obviously that is not going to happen.

Then they started fighting

I found a list of prompts from an engineer, all the steps you need to check. So I ran them, and ChatGPT would read what Manus had said about the solution Manus came up with.

And then ChatGPT started to critique Manus. And Manus started to critique ChatGPT.

They had this fight between themselves, with me in the middle, having no idea what they were fighting about. It was a lot of technical discussion. But it always felt like ChatGPT knew slightly more about what we were talking about.

So I asked it to explain things to me, and I read all the references. And that is where the idea came from: a platform that would teach me about development. I did not want to be the person who did not know anything about what she was doing.

So I built one. For myself.

Which is also something AI gives you, the ability to make things you can actually use. And I think that is what Praximedes is about. Having the possibility of creating things that make sense to you, that you can use for your own self-improvement.

I was sure it was lying to me

At that point I was freaking out that Manus was lying to me. One hundred percent sure of it. And that it was burning my credits while doing it.

I was spending a lot of money building in Manus, because I thought Manus was going to give me my first SaaS. Bear in mind: no knowledge whatsoever. And Manus would suddenly be a sentient being that would tell me, oh, you need to pay attention to security, you need to pay attention to logging your users.

So that was my Manus experience. Me, in the end, one hundred percent certain it was lying to me.

Around the beginning of January I gave up on viral hawk. It was too difficult. I had to pay attention to all these things I had no idea about: security, the vulnerable parts of a product. And no PRD.

I was just throwing ideas, and Manus was trying to make them work together, and I was building the same tool in different places.

Without a PRD, without a direction, you come up with crazy things but you have no path. No roadmap. You are just coming up with ideas from the voices in your head.

Where it went

My Manus experience ended because I did not trust it any more. I moved to Lovable.

And it was also the moment I started seriously considering going back to the idea of having my own app, but not only that. Something else. A layer that would help me get more revenue out of the apps.

That had been in my mind for a while. I had researched it for years, from previous experience. A/B testing, all of that. And I thought: I can build one, with all these apps I already had in Lovable, and add a layer that finds ways to get more revenue out of the subscriptions.

So that was how I moved from Manus to Lovable.

And also the moment I found out what GitHub was about.

If you are building something without a map and you want someone who has already walked the road, that is what my sessions are for.

Techne, episode one. 20 minutes.

This is dedicated to you, the person who is always the most technical person in a room full of non-technical people.

We are in the middle.

It’s funny how life is. You think you have your whole career figured out. You think everything is just not going to be any different from what you envisioned. And then at some point you are working on something that is going to end.

Well, AI came. It’s not ending, exactly. But everything is changing.

For the longest time I thought I had a very clear path, and it was not that.

What AI also did was give me an opportunity to do something I never thought was possible unless I stopped working for a whole year, maybe more, and learned to code again. I had coded a long time ago. Basic PHP, some things I picked up to help me with a job where I was hiring companies and developers to build websites, and I needed to understand what I was buying.

I was, again, the most technical person in a non-technical company. Without being a technical person. So I had to find ways not to get tricked, or not to buy things that were not what we needed. The best route seemed to be to take a coding class, and at that point it was PHP, and I did some very basic things.

Then it got more advanced. That meant actually paying attention and doing the work, and it was real coding, not just programming logic. And I thought: okay, maybe this is not for me.

Sure, we do that

By then I was doing some SEO and I had my own agency. A client came to us and said, the App Store is coming, I want to launch an app, I heard about this thing called ASO.

I said sure, we do it.

I had no idea what that was.

But the internet was about research back then. You just researched. I found out some people were doing this, and I ended up launching my own ASO company later on, and worked in ASO and apps for the longest time.

I always had the itch, though. I always wanted to build my own app. From the very beginning. But I knew it would cost me maybe another year of learning to code, and I was not in the mood for that, and I never had the time. Then later you have kids and life happens and stopping is not possible.

If she can do it

In the beginning I was doing some prompting like everybody else, but I took it seriously. I did the courses. All of them.

Then a friend called me and said, I’m learning to use Lovable, I want to show you what I built.

This is a very non-technical friend. Someone who does not code at all, no technical background whatsoever. And the app was really well built, because she comes from a customer-focused, product-focused background, so she had the feel for building things aimed at having users.

The interesting part is what it did to me. I thought: if she can do it, and I even have some background here, I even know how to open a terminal, I have used an IDE in my life...

Okay. Sure. Let’s try.

Being someone who works better in a team, I got some friends together, found a course on Lovable, and we did it. Two sessions, a master prompt, a whole prototype from the start. ICP, features, learning to talk to the AI so it understood where we were going.

And once again I was the most technical person in the room. Which seems to be my destiny, even though I am not the most technical person I know. It just keeps happening.

So the coding part, the actual writing part, became my part. I read the instructions, I did the sequencing, I wrote the prompts, I built the products.

Lovable at that point was very early. Sometimes you asked for something and it gave you something completely different, and you needed to check what it was doing. If you did not know certain aspects of coding you were lost, because you could not make sense of what came back. And sometimes the prompt was just not clear enough on your part.

December

Then it got to me that I needed to actually go forward with this. AI was good enough now. I did not need a development team. I did not need to code.

At that same moment my life and my plans went to the same place, because I had the opportunity of not being on a project, and I thought that was perfect.

So: December. I had four or five prototypes by then. The two from the course, three more from playing around.

My husband has a channel on YouTube and was always asking me to find things for him. References, videos. I could do it, I just did not want to. So I thought: wouldn’t it be interesting to build a SaaS. Why not build a SaaS. You have AI.

That became viral hawk. Find the references, rank them, put them in a UI he could use, with the analytics in one place. Yes, other tools do that. At that point I thought it was a good idea to build our own.

And we did. Sometimes it worked.

My dynamic was very, very crazy. I would have a conversation in ChatGPT, a folder with projects, everything in there. At that point I had no idea what a PRD was. I mean, I knew what a PRD was. I had never had to build one. I came from companies with developers and product people who took that part, and even though I was in the meetings and knew the basics, I never had to do it myself.

So ChatGPT, with no context, could not help me that much. I did not know how to specify things, so the tool came up with its own solutions. Sometimes Manus was on point and sometimes it was not, and at some points it felt like it was playing with me. Session after session, spec after spec.

The bug

At some point I wanted to publish a website, my own consulting site, which I wanted to look good, and I could do it myself now.

I could not publish it.

It turned out that if you did not put the https://www. in front, the Manus platform would not show your site. Something as simple as that. You could not just use your domain. It had to be the whole thing.

It was weeks. Weeks of discussions with the Manus people, and they could not help, they did not know how to help me.

And then one day, by accident, I figured it out.

Manus was a stepping stone to understanding a lot of things. Manus was also a bit of a challenge.

That was around December. I was still learning what to do with a PRD. I didn’t know how to build one. In fact I did have a GitHub account, by the way, but I had no idea why I had it.

And December was the start of the shift that turned me, completely by accident, into a terminal person.

There is more Manus coming. But these will be 20-minute drops, so you can listen while you’re driving, commuting, or just trying to listen to someone talking about life, the universe, and a bit of tech mishaps.

If you are building something yourself and you have lost weeks to a wall like that one, that is what my sessions are for.