April Newsletter

 1. Those who are not married try different partners

Some of us are "married" to a specific chatbot, usually because it's paid for by the company we work for. But for everyone else, it's a time of great temptations. "I'm bigamous," a friend admitted on Monday, confessing to using both ChatGPT and Gemini. A few days earlier, a colleague showed me a corporate simulator created with Claude, adopted after months of using OpenAI models, to explain to engineering students the effects of certain choices.

This trend of abandoning ChatGPT or supplementing it with other models is not only evident in my experience but is confirmed by recent market analyses (one, two, three). So, it's necessary to understand whether generative AI services should be treated like traditional utilities, such as electricity or telephony, where you just choose the best provider at the moment, or like streaming services, for which we've accumulated multiple subscriptions over time. It's interesting that OpenAI's loss of market share has not even slightly hindered its growth: it continues to have the most-used AI models in the world and has just completed a funding round of (another) $122 billion.

My personal experience is that today I would almost exclusively use Claude, but the offer it proposes is not flat, and every two days it asks me to recharge my credit. This pay-as-you-go offer complicates shared access to these resources, as many people do today.

2. Developers, tough times ahead

Agriculture, which at the beginning of the last century represented the core of employment, underwent a drastic numerical contraction not because we stopped eating, but because we became incredibly efficient at producing food with a fraction of the initial workforce. In fact, today we produce much more food than we did a century ago. The image below illustrates how automation and mechanization transformed a labor-intensive sector into a super-productive one with reduced use of people.

Today, observing the evolution of programming models and assistants, it's hard not to foresee a similar fate for the software development sector. The recent AI Index 2026 report from Stanford University confirms that the workforce disruption caused by artificial intelligence is no longer a forecast but a tangible reality that is primarily affecting new market entrants. We have entered an era where code has become a commodity, and tasks that until recently required hours of manual work by junior IT professionals are now handled more efficiently by autonomous agents.

This transition is supported by very recent macroeconomic analyses. A BCG report published just this April 2026 estimates that over 50% of professional roles will be reshaped by AI within the next two years. In software, this translates into the collapse of entry-level positions, which in some regions have seen a drop in hiring exceeding 45%. The economic logic has changed: it's no longer cost-effective to pay a human resource to write unit tests or boilerplate when a senior engineer enhanced by agent tools can produce the volume of work that in 2020 would have required an entire team of junior developers.

What makes me smile is that it was developers who created the AI that is now eliminating developer jobs.

The Society of Authors (SoA), the UK's main trade association and union for writers, illustrators, and literary translators, thought it would be useful to create a label certifying whether a work was created without the use of AI. If once AI pretended to be human, now the burden of proof is reversed, and we must demonstrate that cultural products were not generated by AI. This proves that the Turing test is completely outdated, and we wouldn't be able to distinguish who generated content; thus, just like an organic product whose production chain we can't personally verify, we need a "Human Authored" certification.

Personally, I find it disturbing to think there's value in not using AI. If I write a book with AI and make someone cry, is that emotion less real? Perhaps the label can prove that behind the text we buy is a soul, and this characteristic should testify to a potential "connection" with the creator of the work, but how many books from the past have we read finding them empty and useless? The assumption that the real person is by definition better than AI is hypocritical, just as much as thinking "natural" is synonymous with "good," or that the pictorial representation of a subject is always more valuable than its photograph. I think it depends on who uses the brush or the camera.

This approach reminds me of the country-of-origin bias, where a Made in Germany product is reliable and a Made in Italy product is beautiful, or the genetic fallacy, where an assertion is accepted or rejected based solely on its origin or history, ignoring its intrinsic merit.

7. Where is my mind, Mythos?

Anthropic has announced an AI model (called Mythos) capable of discovering cybersecurity flaws never uncovered in decades of attempts. The system is so powerful that it has brought together companies like Amazon, Apple, Google, Microsoft, NVIDIA, JPMorgan Chase, CrowdStrike, and others for the Glasswing Project, where they can secretly test their Mythos, securing the entire world from cyberattacks.

It seems evident that this is a super-weapon that, if it fell into the wrong hands, would cause enormous damage to humanity. It feels like the plot of an apocalyptic movie, where what was supposed to remain locked in a laboratory escapes the hands of researchers, and only the US can save us from global destruction. The Glasswing Project aims to give an advantage to those who have developed software, allowing them to discover problems before this analytical power reaches hostile individuals. This time, I don't feel very reassured, and I already hear the Pixies in the background as buildings collapse.