I want to make it clear that this is the opinion of someone who is genuinely passionate about building applications.

I have been building websites, web apps, mobile apps, servers and so on since 2015, more than 10 years.

In all that time I have been through everything: tooling changes, constant updates, problems with no solution for months on end.

In the last three years, however, the paradigm has changed completely.

Before the 2022 release of ChatGPT that turned the world upside down, we developers had several options for solving a problem.

We would search Stack Overflow first. If we found it quickly, great; if not, we had to read the documentation for the tool or service.

If we got stuck, we had two more options: post a question on Stack Overflow, or ask someone with more experience than us directly (the wise ones).

Once ChatGPT arrived, the first option was relegated to asking the chat. Fine so far. A new tool that spared us searching over and over and digging deep into Stack Overflow. A saving in time, but not in creativity.

After that came new plugins and tools, like Cursor, which embedded the ChatGPT chat, or other models, and also gave you AI autocomplete on tab. Same as before: same creativity, faster execution.

A few months ago coding agents appeared that practically program for you. Claude Code, for instance. This marvel, or nightmare, depending on how you want to see it, can write 99.9% of your code today.

Many companies still restrict their developers from using AI and, in my opinion, they are shooting themselves in the foot, because they are competing against companies whose development is vastly faster.

A large part of the market believes the code introduces security flaws, spaghetti code, bugs and so on. All of that is true if you do not have a good copilot, meaning someone with technical knowledge monitoring what the AI does, plus solid documentation of both style and architecture rules.

Either way, what is being asked for today is speed. Teams are shrinking. Developers do more with less. But how do we developers feel about it?

Yes, I will say it again: to me software development is dead. Even the software workflow is broken. You used to start with planning and then a requirements analysis. Today you start with the requirements and Claude does the planning for you.

Beyond that, what saddens me most personally is the loss of creativity. Every developer used to have their own style. To me, programming was a form of art. It was devising, through ingenuity, a solution that had to be clear, simple, optimal, concise, secure and maintainable, and every developer got to the same place by a different road.

Today it means delegating everything that truly fascinated me about this discipline to Claude. You wait ten minutes for the AI to hand you the solution. You read it, refine it, read, refine, read, refine.

Can you imagine a doctor who never again has contact with their patients? A robot does it and the doctor only watches a recording of the key moments. Or a lorry driver whose job is to supervise the self-driving truck and only watch the road. Would those not become boring jobs?

Software development has lost its soul. It no longer carries emotion, and it barely makes you think. These days I find myself looking for alternatives to feel creative again. But we will never feel that way building software again.

If you are in the same position as me, I would appreciate you leaving your opinion. I would understand it whether you agree or not. It is true that we have gained in ease and convenience, but we have lost in everything else.

The opinion of someone who has been developing for more than 10 years and feels this discipline has lost its soul.

Software development is dead

YOU ARE WASTING TIME. While you get on with more important things, you can build applications, FOR FREE, and with no technical knowledge. Just an idea. And you can have FULL CONTROL from your phone 🤯

Right at the bottom I leave you an example of the LATEST thing I built with this setup, while doing other, more important things 👇

In fact, for simpler things with no server, Vercel alone is enough (you need neither MongoDB nor Railway). I prefer this stack because latency to the database is lower. That is, if your application lives on Vercel with NextJS and you have both the frontend and the server there, every database request has to open the connection, which adds latency.

1. First we create our repository, which is where your code will live. Go to github.com, sign up and create a repository.

2. Get Codex. Go to chatgpt.com/codex/pricing and buy whichever package suits you. You also have the option of Claude, which personally I like more, but it is pricier (the PRO plan starts at $100). At this point we are getting the AI subscription that will generate our app's code.

3. Install OpenClaw (my pick) or Hermes Agent on your computer. This gives us an agent running 24/7, connected to your phone, which will handle basically everything. The idea is to keep it on a machine that stays on all day. You have several options: install it on a Mac Mini (the best), install it on a VPS (less control but cheaper), or THE BEST FOR BEGINNERS: install it on your own computer and leave it on all the time. This is OpenClaw's official site: openclaw.ai. You install it with the following commands:

Windows: powershell -c "irm https://openclaw.ai/install.ps1 | iex"
Linux or MacOS: curl -fsSL https://openclaw.ai/install.sh | bash
Npm: npm i -g openclaw

Once installed, run: openclaw onboard. Follow the onboarding. Here you should add Codex/Claude at the "model selection" step and link it to your phone, whether through WhatsApp/Telegram (the best), Discord or Slack. Once you reach this point and can send a "hello" from your phone and have OpenClaw reply, it is important to give it access to your Github repository. IMPORTANT: create a folder on the Desktop called "App". Tell OpenClaw to always work in the "App" folder on the Desktop.

Generate a repository called [APP_NAME] in the "App" folder on the Desktop and initialise it with a 'Hello World'. It is important that you build a monorepo with yarn workspaces where one folder is the frontend and another is the server. The frontend will be built in NextJS and the backend in NodeJS with Fastify. I am going to connect the server to a MongoDB database. Generate a .env and create the MONGO_URL environment variable for the MongoDB connection. When you are done, make the first commit on the main branch.

5. Once OpenClaw finishes, go to Vercel. Create an account using the Github login/sign up. Your repository will show up. Select that it is a NextJS project, with the root set to "frontend". This will deploy your application's client to a Vercel subdomain (xxxxx.vercel.app). Later you can point whatever domain you want at it.

6. Go to mongodb.com and create a Mongo Atlas account. This will be your free database. Create an organisation, create a project and create an M0 cluster (free). Once the cluster is created, hit the connect button and get the connection string (it will ask you to create an "internal" account, create it). The string will look something like this:

mongodb+srv://account_name:account_password.cluster0.xxxx.mongodb.net

7. Go into the App folder on your Desktop. You will see your repository there, with the name you gave it. Inside you will see the "frontend" and "server" folders. Open server, find the ".env" file and paste that connection into the part that says MONGO_URL. It is very important that you know where this file is, because it holds the sensitive information and secrets your server will use. In the future you might add payments to your app and you will need to put the STRIPE API_KEY in there, among many other possibilities.

8. Go to Railway, create an account, log in or sign up with Github and link the /server folder. Copy and paste the .env file into the "variables" section.

At this point your application is already live at xxxxx.vercel.app, the subdomain Vercel generated for you. That is where you will always see the result. Now open your WhatsApp/Telegram and tell your OpenClaw to run your idea. In a while you will have it live on that Vercel page.

You will be able to build games, dashboards, forms, store information: practically anything that crosses your mind. Meanwhile you work, eat, go to the gym or sleep. The agent will do it for you, and it will always be available on that web page. You do not need technical knowledge, just follow the steps to configure the setup and you can build anything.

There are two points left to cover in all this, but they matter for finishing a complete application: payment handling (Stripe) and mailing (Resend). With both it is as simple as signing up and getting the following:

With that same setup I built the app you will find at billionaireswar.com. It is basically a game where people sign up with their social accounts and buy countries so their avatar shows on top of each one (a way of promoting yourself, "The Million Dollar Page" style).

The complete setup for building applications for free, with no technical knowledge and full control from your phone.

Build apps while you eat, work or sleep

AI is not intelligent, that is what you have been led to believe. The big companies sell you AI as something with superpowers, but it is all a lie. If you genuinely think it is intelligent, I invite you to read this article in full.

AI does not think: it makes mathematical predictions

If you believe AI thinks and reflects like a human, you are very much mistaken. Basically, and to put it plainly, all it does is predict the next word or pixel based on the previous words or pixels (the context). An AI is trained on enormous amounts of data and, through mathematical models, learns patterns and calculates probabilities to predict which outcome is most likely in each situation.

For example, imagine someone completing sentences: "It is very ___ today". Your brain would probably say "hot". AI does something similar, but without understanding anything. It calculates probabilities:

It picks the most likely one and keeps doing that over and over. Basically it finds: in text, the next word; in images, the next pixel; in video, the next frame; in music, the next sound. It all comes down to enormous mathematical calculations to predict what comes next. Language models are trained precisely to predict the next token (chunks of words, or words).

Chess is the best example there is for showing that AI does not reason. If we wanted to snapshot every possible game of chess we are talking about 10¹²⁰ games. There are 10⁸⁰ atoms in the universe, to give you a sense of the magnitude. That means there are unexplored games and positions, and that many games are not inside the datasets the AI was trained on.

So some researchers set language models to play chess. At first they seemed to understand the game perfectly: they played known openings and reasonable moves. But in certain situations they started doing absurd things: illegal moves, pieces passing through other pieces, impossible rules. The problem was not that they had never read the rules; the rules were in the training data. The problem is that recognising movement patterns in games the model had not been trained on produced errors. While the positions resembled known examples they worked well; when the situation drifted from those examples, the AI failed and made illegal moves.

It is like a student who memorises thousands of exams: they look like a genius until you change the questions. Paper: openreview.net/forum?id=65R1Dbfwzk

It has no ideas of its own: it mixes existing ones

It is the opposite of being creative. You think AI is creative because it makes you a nice design you have never seen before, or a web page with an unprecedented interface. But what it is really making is a collage of components that already exist and are inside the dataset, which together look like something "new". This tends to homogenise results.

If you built 100 web pages using AI, you would notice it repeats the same patterns, the same transitions, the same button appearance. That makes nothing original, and if you are applying AI across every area of your business, you are killing originality and drifting towards looking just like the competition.

There are headlines about new proteins being found or new medical cures being discovered. Is that actually false? Honestly, no. It is true that AI helps "discover" new things, but not because it was creative and generated something unprecedented. Rather, those patterns or data were already there, and humans had overlooked finding and identifying them.

The difference is that an AI does not have inspiration the way a person does. It does not wake up one morning and think: "What if I try this because I have a hunch?". What it does is analyse absurdly large amounts of information and find relationships that are practically impossible for us to spot. Where a researcher can review thousands of possibilities, an AI can review millions.

It is like looking for a needle in a gigantic haystack. The needle was always there; the AI did not create it. What it did was find patterns and connections hidden among enormous amounts of data.

What is genuinely revolutionary: a gigantic shared memory

This may be where the real revolution is. Not an artificial intelligence, but a centralised artificial memory. Imagine a library holding almost every book, almost every article, millions of conversations, images, videos, code and research.

We already had this with the internet, but all the information was "unsorted". We had to find it ourselves. Now we have all of it centralised, from a single chat, in seconds, and it can also find relationships across all that data. That really does change the world.

Should we call it "Centralised Artificial Memory"?

Perhaps a more honest name would be "Centralised Artificial Memory". Because it better describes what it does: store human knowledge, relate concepts, find patterns and return useful answers.

The word "intelligence" makes you think of consciousness, understanding, intent, independent thought and creativity. And that is infinitely far from what we currently call artificial intelligence.

AI is not intelligent, that is what you have been led to believe. It does not think: it makes mathematical predictions.

Artificial intelligence is a scam

Something is happening that few people are talking about, and you should be taking advantage of it right now. Think of it as a lunar eclipse: instead of the Sun we have artificial intelligence, and instead of the Moon, social media. Let me walk you through it step by step:

AI today helps you iterate and pivot fast. You have an idea, you ship it, you check whether it worked, you improve it, and you start again. If you fail, you drop the idea and go for another. Completing one of those cycles used to take months. Now it takes days. Development speed has multiplied, and the trend is not linear but logarithmic.

On top of that, and this matters a lot, AI is still cheap for how useful it is. My personal view is that we are at the equator between useful and cheap. For €100–200 a month you get a ChatGPT (Codex) or Claude subscription that lets you run agents in parallel and do several things at once: writing code, research, planning, documentation, testing, and in any area of the business.

The companies behind AI are losing millions making all this so accessible. It suits them too, since every user helps improve the models with our responses, and they are getting us used to living alongside AI. But sooner or later this will end.

The trend is already becoming clear: access to frontier models will get more limited and more expensive. Claude Fable 5, for example, will stop being available inside the Claude subscription from 7 July and will only be reachable through the API, pay as you go. In practice that can send usage costs through the roof. I am not saying it is imminent, nor that it will happen for certain, it is just something to bear in mind.

In short: AI is very cheap (for now), it is very useful and it makes you infinitely more productive.

Social media is the biggest shop window for people to see what you do, and it is free on top. Before I carry on, a message for anyone reading who feels embarrassed about publishing content: nobody cares about you. The first day you post something, not even your partner will see it. So post. Share what you know and add value. You will not regret it.

That said, and above all, if you have a business and you are not posting daily on every social network, you are wasting time. A few months ago the main social networks changed their algorithms. YouTube, Instagram, Twitter, LinkedIn, TikTok, Threads: all of them! They used to reward accounts with more followers. Starting out was hard. Now it is very easy to post content with zero followers and have something go viral. Every social network is rewarding it. Take advantage of it!

Let me give you a few examples, since I am practically starting out on all of them. On Twitter I began three months ago. I have posted every day. Almost 30,000 followers.

On Threads I started two days ago and I have already got more than 1.2M views.

On LinkedIn, 60,000 impressions in two days. After making content months ago and getting practically nothing.

On YouTube, Instagram and TikTok I have only just started (I have a single video), so I have no meaningful stats yet. And really, none of this is magic. The strategy is very simple: I combine engagement posts (50%) with personal posts (50%). Examples:

Engagement posts are basically me seeing what works in my niche, copying it and applying my own style. That is all. It will go viral, especially at the beginning. Knowledge posts reach far fewer people, but they build the authority engagement posts do not. Aiming to polarise helps too (Grok is very good for that).

The thing is, with a single click you make content for Twitter, Threads and LinkedIn. Exactly the same content. The same goes for vertical videos, for Instagram, TikTok and YouTube Shorts. And while you add value, you can get known and promote anything, for free. Channel those millions of impressions and views into your products and businesses. People do not just see your post. They go into your profile, look at who you are and what you do.

I will say it again, and even more bluntly: if you have or want to start a business, and you are not posting on Instagram, YouTube Shorts, TikTok, Twitter, Threads or LinkedIn, you are wasting the opportunity of your life.

Put those three together and you are at a better moment than $BTC in 2012 to grow, feel useful, add value to society and, of course, make money.

Cheap, useful AI and algorithms that reward the small player. Put the two together and you have the opportunity of your life.

We are at a $BTC-in-2012 moment and you are wasting it

Training is the biggest pillar of my routine. It is what gives me more energy every day.

It is my only way to switch off from everything. I just focus on getting the ball back.

Padel

I tend to read founder biographies, they inspire me. When I founded Different Health I was right in the middle of Phil Knight's «Shoe Dog», and around then we started working with Nike. That was a lot of fun.

Reading

Trying out tools and building applications with no goal beyond seeing how far they go.

Side projects

This is what I enjoy doing once the computer is off.

Universitat Autònoma de Barcelona, class of 2019. The foundation everything else rests on.

BSc in Computer Engineering (UAB)

Honours in Software Engineering: architecture, quality and the craft of building systems that hold up.

Specialised in Software (university honours)

Products, automations and systems with artificial intelligence applied to real business problems.

AI expert

Threat modelling and secure development as Security Champion at IriusRisk.

Cybersecurity training

Computers, swimming and guitar

From Telecoms to Computer Engineering

My first business

From the big company to startups

Different Health

Taking distribution into my own hands

40,000 people and an AI academy

Twenty-nine years, in short.

Fran Pradas,

engineer and founder.

Essays on artificial intelligence, software development, and how products and audiences get built.

What I write about

AI and product.

More than 40,000 people follow my content on AI across X, Threads, LinkedIn, Instagram and TikTok. I share tools, lessons and practical cases.

We improve our clients' health by running a battery of tests such as VO2 Max, strength, balance, bone analysis and blood work. I run the engineering and product teams and the health compliance side, including HIPAA.

Technical leadership to improve the platform and the services affecting the elderly population across the Nordics and Central Europe.

Threat modelling to design and build secure applications proactively. Migration of the client layer to modern tooling.

Internal applications for Mango's day-to-day operations: time tracking, an incident portal and real-time market trend analysis.

career

Fran Pradas' companies and products

Investment banking, retail, sport and the public sector.

Companies I have

worked with

Eight years in the profession

Startups, products and a community that grows every week.

Social media

Write to me and I answer in person. No endless forms, no automated replies.

Shall we talk

this week?

I turn ideas

into digital products

I design systems so products find their audience through social media, content and UGC.

I share tools, lessons and practical cases to help more people use artificial intelligence in a useful way.

We are building a new way to understand preventive health through data, technology and personalised experiences. Co-founder & CTO.

I explore how to connect artificial intelligence, design, automation and distribution to build more useful and efficient products.

We design and build bespoke applications and websites, caring for product, experience, technology and brand.

I work at the intersection of product, technology, distribution and community.

What I am focused on

right now.

The page you are looking for does not exist or has moved.

Page not found

I am passionate

about technology.

Preventive health platform. Advanced analytics, wearables and personalised plans to live longer and better.

Web design and development studio. Send us the quote you already have and we improve on it: more reach, a better price, or both.

We Make Studio

The AI academy for absolute beginners. 52 fifteen-minute lessons, in plain language and with a teacher who answers.