This post has been originally published in the Fanzine Digital Soverignty which you can read for free here: https://blogs.amarinha.gal Be aware that the text is published in Galician there.

Before I start writing this article, I want to propose a game to the readers. Since this piece will deal with artificial intelligence and its applications, I would like to ask you to try to identify whether this article has been written by a real person or by artificial intelligence. I will come back to the answer at the end.

One of the questions I receive most often in my day-to-day as a tech lead is when the “machines” will replace programmers in their jobs. Depending on the person asking the question, the term machines will be substituted by others like “AI”, “machine learning”, “large language models” or (if the asker is especially deep into the topic) even “agents”. In the end, the question is clear and the worry is real: we have come to a turning point where computer programs can and will take over a big part of the menial tasks many of us currently get paid to do.

As my crystal ball is currently in the garage for maintenance, my answer in these cases is rather vague. It usually goes along the lines of “I am not sure when some jobs will be fully replaced, but whoever does not know how to leverage AI will be left behind.” Regarding this point, I am rather certain, as it is happening even as I write.

There are a couple of caveats that I would like to mention before being flagged as a catastrophic futurologist. The so-called AI revolution will affect, in the first steps, only digital endeavours. This is the case due to the relative underdevelopment of robotics, which, as the Terminator movies showed viewers decades ago, go hand in hand when it comes to worrisome future prospects. Robots (that is, the embodiment of the AI algorithms) are lagging behind in terms of capabilities, and that should grant a sense of security to all those that “need to move” in order to accomplish their jobs.

The second caveat is commonly referred to as replacement theory. Even with the latest advancements in terms of “agentic workflows empowered by MCP and last generation attention transformers” (or whatever else the latest keywords have come to be), there is still real replacement in terms of new job positions around this fascinating new tech, which allows many digital workers to find new niches that simply did not exist four years ago when chatbots were at the level of a 5-year-old.

Now, I do believe that bluntly ignoring AI will render some people simply unemployable, as the productivity gains in certain areas are just too big to ignore. In a previous edition of this fanzine, the concept of Wikipedia and its many related sites was discussed. These are tools without which our lives would have been worse. In a similar fashion, I am of the opinion that AI and specifically LLMs allow us to accomplish certain tasks that before were previously accomplished in an ad-hoc manner or not at all. Translations, text summarization, text generation, knowledge access and many other tasks are now one “prompt” away, conveniently packed for immediate consumption.

When it comes to digital sovereignty, these models are indeed a nightmare. The computational power required makes the top algorithms almost impossible to run on a local machine like your laptop or phone. This means that we, as consumers of these tools, need to either run a watered-down open source version locally (prepared for much less powerful hardware) or send the prompts (our data) to the cloud, where it can and will be used by the corporation running the system—first to train the algorithm further, and (not much) later to resell this data to any party interested. Rest assured, any data already made public before the advent of these new algorithms has already been used for this purpose. Heck, it has been proved that all kinds of copyrighted work have already been used—from books, to movies, to illustrations and, of course, all the pictures that we have been gladly donating to Facebook, Google and co. over the years.

As we turn more and more to Grok, ChatGPT, Llama, DeepSeek, Copilot and friends, we need to be conscious that even though we are not talking to a real person, our data flows to the engineers behind. Being sensible and not oversharing is the modern equivalent of “do not tell your chatbot what you would not share with the Alexa sitting idly in your living room.”

Lastly, if you were wondering whether the article has been written by a real person or an AI, rest assured, it has been both. The original text was written in English by me, but grammatically corrected and fully translated to Galician by an AI.

Glossary

• Artificial Intelligence (AI) is the characteristic of systems to show human-like behaviour.

• Machine Learning (ML) is a somewhat similar term, but currently refers to “teaching” a machine how to accomplish a task based on examples.

• Large Language Models (LLM) are the current top-of-the-class algorithms when it comes to text tasks. The well-known ChatGPT is a kind of LLM.

• Prompts are each of the chats we send to an LLM.

• Model Context Protocol (MCP) is the interface that allows LLMs to communicate with other “more traditional” software like email, browsers, office suites, etc.