software

so that the software still fits on a floppy disk

by

We have become accustomed to software weighing more and more. We see it in applications that take time to download, in simple tools that come with too many layers and in services that promise convenience in exchange for take up more spaceconsume more resources and depend on more invisible parts. That is why it is striking that, in 2026, when much of the technological conversation revolves around AI and increasingly ambitious systems, there are those who claim an idea that seems to have come from another era. The initiative is called Fits on a Floppy and part of a manifesto published by developer Matt Sephton. Its rule is as simple as it is striking: an application that wants to show off its badge must have a total download size of less than 1.44 MB, the capacity of a classic 3.5-inch floppy disk. The text itself summarizes it with a direct phrase, “the software has lost its way”, but its proposal is not to miss the physical support, but to recover the discipline that was imposed by working within very narrow limits. For a long time, making software was also about giving up. If something was not necessary, was left outbecause memory, storage and user patience had a very visible limit. Then came a different stage: the teams they began to have more margindownloads stopped feeling like an adventure, and the size of an app stopped being a central concern. There a dangerous door began to open. The software has not gained weight by accident Not all of that growth came from adding visible features. Much of it came underneath, in the form of layers that the user does not always see: libraries, engines, update systems, components designed to support more than one version of the same product and dependencies that allow progress to be made faster without solving each problem from scratch. That way of building makes sense in many cases, especially when you want to maintain the same product in several systems. But the scale also changes. This is where the real value of Sephton’s proposal comes in. Fits on a Floppy is not trying to show that everything should be compressed to fit into 1.44 MB, but rather that an artificial restriction can serve to prioritize. If an app is born to solve a specific task, the manifesto asks that it download quickly, start without waiting, consume few resources, be native and avoid unnecessary dependencies. The underlying idea is simple: the less baggage a tool carries, the easier it is to understand what it does, why it does it, and how much it costs to maintain it. The question, then, is whether this discipline can once again have a journey outside the manifesto. In some software, probably yes. We are not talking about browsers, video editors or services with integrated artificial intelligence, but rather small utilities, single-function tools and native applications that often do not need to carry a huge architecture. There Sephton’s argument is stronger: if the objective is limited, the size should also be limited. Not out of nostalgia, but because a simple tool has fewer excuses to behave like a full platform. The other side of the story is that much of the software is not going to get smaller. Many current applications are no longer just a window with a specific function: they integrate accounts, synchronize data, offer real-time collaboration, work on multiple systems and they accumulate functions that were not part of a desktop application years ago. All of this may be justified, but it weighs. That’s why the promise of returning to lightweight software has clear limits. For many products, the real question will not be whether they can fit on a floppy disk, but whether they are growing out of necessity or accumulation. The beauty of the floppy disk, in fact, is that it no longer seems reasonable. Precisely for this reason it forces us to look at the software from another place and ask ourselves if all that weight responds to a real need or to an accumulation that no one dared to review. Fits on a Floppy is not about stopping the evolution of modern tools or denying that many need to be big. Its usefulness lies elsewhere: reminding us that efficiency is also a design decision, and that the size of an application says something about how it was intended. Images | Fernando Lavin In Xataka | iOS 27 doesn’t leave any iPhone behind, but WatchOS 27 can’t say the same with Apple Watches

Software engineers have become AI overseers. Another profession is going the same way: lawyers

by

AI already writes faster than many professionals, making it a powerful ally in administrative and legal textsthus lightening the most tedious workload for humans. The problem is that it is an efficient tool, but not infallibleso supervision of its results is necessary. That is, it saves you only part of the work, but maintains the obligation to monitor the results. That is just what the TSXG has put on the table in one case which affects a Galician lawyer, against whom the Galician High Court has opened a case for procedural bad faith after detecting a written appeal with AI in which he had invented 24 legal quotes to argue it. The resource under suspicion. The Superior Court of Xustiza of Galicia maintains that the document presented “multiple spurious citations, non-existent resolutions, others that have nothing to do with what was discussed, or directly invented.” According to the Galician court, the appeal accumulated a series of errors that cannot be attributed to human error, but rather respond “to a notable lack of diligence in professional performance.” The resolution adds that the text followed a “curious structure in the way it was written”, something that the court relates to the use of free generative AI. Although, above all, the TSXG disgraces the lawyer for the lack of subsequent verification to stop the “constant ‘hallucinations’” of the system. Lack of review. As the Galician court’s brief highlights, the procedural bad faith of which the lawyer is accused does not lead to a malicious use of deception, but rather “evidence conduct revealing obvious negligence on the part of someone who, considered an expert in procedural rules and respectful of the deontological principles of his profession, entrusted his work – to what we believe -, without further review, to what the algorithm proposed, omitting the diligence of verifying the existence of what he cited, trusting perhaps in which the abundance of references would not only go unnoticed by this Court, but would also give authority to its assertions.” That is, it is a warning to navigators that AI can speed up writing, but it does not know if a sentence exists or if a decontextualized quote serves to support a legal argument. That process of screening and supervision is still human, and that role of supervising results will be a key piece in the future of the practice of law. Engineers have already changed. The process of changing roles that the legal profession will undergo in the very near future (in fact, as we see, it has already begun) with the arrival of AI is not something new. Software engineers and programmers were the first to experience it. AI is gaining prominence when writing the code of large technology companies like googleMicrosoft and even Anthropic already generates much of its AI with its AI But, at the same time, the engineer has not disappeared, but has gone from having a role as a producer to one of a supervisor who analyzes the results provided by the AI. By that same reasonthe technological have reduced the hiring of junior personnel, who until now were dedicated to those more tedious tasks, and have accelerated the hiring of seniors with the most trained eye for detect errors and inconsistencies in the code. From legal advisor to legal supervisor. The comparison between what is already evident in big technology companies and what is coming in law firms is quite evident. AI is not going to end the figure of the lawyer, but it will generate a shift in work roles towards the supervision of legal texts. In fact, as how I collected Reuterssome law firms are already requiring their junior associates to know how to use and supervise AI. We are only facing one of the cases derived from the improper use of AI in legal texts that have been occurring in recent months. According what was published by Iberley In February, the Superior Court of Justice of the Canary Islands established a financial penalty for a lawyer who had used AI to generate an appeal that contained a total of 48 jurisprudential citations. invented by an AI. Regulate the use of AI for judges. In reality, the impact of AI in the legal field goes beyond its use in drafting routine legal resources and briefs. In Spain, judges have already been sanctioned for using ChatGPT in drafting something as compromised as a sentence. In April 2026, the General Council of the Judiciary (CGPJ) opened a file to a magistrate for having “evaded his jurisdictional functions” by leaving the drafting of a sentence in the hands of an AI. As a result of this case and other controversies, the General Council of the Judiciary has had to approve a “guide” aimed at the use of AI in the judicial career, so that it is established how this tool should be used so as not to incur a neglect of one’s legal responsibilities. Who knows if one of those new professions to be created in the future it will be that of supervisor of sentences. In Xataka | “Unpresentable” and meaningless: the court declares a dismissal unfair due to a letter generated with an AI Image | Unsplash (Romain Dancre and Solen Feyissa)

The ‘vibe coding’ promised to democratize software. Your first gift is 5,000 apps with open sensitive data

by

An investigation by the firm RedAccess has found more than 5,000 applications created with tools vibe coding which practically lack authentication. Anyone who stumbles upon its URL can enter. Of those 5,000, 2,000 appeared to contain private data upon inspection. The finding covers apps generated with Lovable, Replit, Base44 and Netlify, four of the platforms that have most popularized describing a program with words and letting a LLM write it. Why is it important. The promise of vibe coding is that anyone, without knowing how to program, can build software. The catch is that this same “anyone” also doesn’t know what questions to ask an application before releasing it on the Internet. The result is a new category of breaches caused not by careless employees or advanced attackers, but by people who have thrown together an internal tool in an afternoon without going through anyone on the security team. In detail. Researchers have located these applications by doing normal searches on Google and Bing, combining the domains of each platform with generic terms. Nothing of hacking: It’s more like reverse engineering a search engine. What appeared behind those URLs included hospital quadrants with doctor data, company strategy presentations, complete records of chatbot conversations with customers (with names and telephone numbers), and freight books from transport companies. In some cases, access even allowed them to gain administrator privileges and expel others. Between the lines. The platforms involved have responded with the predictable argument: it is the user’s fault. Replit remembers that its apps can be marked as private with one click. Base44 maintains that its access controls are robust and that disabling them is a conscious decision. Lovable points out that its role is to provide tools, not configure them for anyone. It is a valid argument and, above all, comfortable. It is also the same one that Amazon used with the buckets Misconfigured S3 leaking Verizon data or from WWE: the setting was there, but the user didn’t find it. The context. He vibe coding takes an old problem to a new level. Every time a layer of abstraction has democratized a craft (like spreadsheets, the wrappers of AI or web templates), the newly arrived group has arrived without the baggage of good practices that the previous one had. What changes now is the speed. Someone from a non-technical department can create a tool in two minutes and upload it to production without it going through IT. Yes, but. The AI ​​models that generate the code are not neutral agents. They do what is asked of them, no more, no less. If no one tells them “protect this in X way and implement Y,” they won’t do it. Security by default is still not a learned behavior in most of these tools, and that is a design decision of the platforms, not the end user. The consequence is foreseeable. There are going to be many more leaks like the ones RedAccess has caught before the industry internalizes that a “publish” button should not coexist with a privacy setting hidden three menus below. In Xataka | I have lived the “miracle” of vibe coding: this is how I programmed an Android TV app without having any idea about programming Featured image | Xataka

The US has invested 16 years and 8 billion dollars in renewing the software of its GPS network. Result: a failure of epic proportions

by

The Next-Generation Operational Control System project (OCX) was going to modernize the constellation of the United States’ more than 30 GPS satellites. The company RTX Corporation (previously known as Raytheon) managed to win the project in 2010 with a budget of 3.7 billion dollars. The project was supposed to be completed in 2016, but in reality the US has spent $8 billion and 16 years later has an absolute disaster on its hands. 16 years of broken promises. In 2010 the iPad had just appeared on the scene and cloud computing was a somewhat diffuse concept. The project of the US Government was reasonable, and proposed that the OCX system be operational by the time Lockheed Martin’s new GPS III satellites debuted. The development became a chaos of bugs and requirements changes, and to this day it is unclear when, if ever, it will be completed. In Xataka 90% of Iran’s oil industry depends on a tiny island. One that is already on the radar of the US and Israel A fortune invested. The financial management of the project is the first big disaster. The initial budget was estimated at 1.5 billion dollars, but since the award until today that figure has risen to reach almost 7.7 billion of current dollars, to which another 400 million are added to support an improved version of the satellites, the GPS IIIF. This increase is not due in large part to the project suddenly being much more ambitious or more capable, but rather to the costs of having been fixing everything that has gone wrong since they started working on it. Software costs more than satellites. Every time software fails an integration test, the bill runs into tens or hundreds of millions of dollars. That has made the OCX system one of the most expensive and least efficient software projects in recent US military history. In fact, it far exceeds the cost of the satellites themselves that it had to control: the 22 GPS III satellites of the contract signed in 2018 have a budget of 7.2 billion dollars. Satellites of the future controlled by a fairground shotgun. Currently the United States has a fleet of GPS III satellites in orbit capable of emitting much more powerful “M-code” signals and interference resistantsomething that among other things allocates them especially for military applications. The problem is that since the OCX software not workingthey are managing them with control systems inherited from the 90s. It is as if we had a VHS video connected to watch movies on an 8K Smart TV: the potential is there, but one of the components is an absolute bottleneck. {“videoId”:”x8wlh9q”,”autoplay”:false,”title”:”United States vs. China: The CHIPS WAR”, “tag”:”webedia-prod”, “duration”:”1611″} The cybersecurity nightmare. One of the big problems of this project has been the cybersecurity requirements. OCX was designed to resist cyberattacks from powers such as Russia or China, but that requirement has become a spectacular technical burden. Pentagon standards have evolved so quickly that they have not been able to be adapted to an architecture that begins to become obsoleteand covering successive patches is locking the system in a complex vicious circle: the software is never finished because more and more vulnerabilities appear. Failed tests. The latest report from the Government Accountability Office (GAO) has been the final straw. During the tests the system again showed once again instabilitywhich has forced the final delivery to be delayed to the end of 2026 or even 2027. Frank Calvelli, of the Air Force, has expressed his dissatisfaction with that unacceptable management of private industry: the strategic advantage that this project should offer at a time like this is inaccessible due to the disastrous progress of the project. It’s not that difficult. for a long time the excuse for justify the delays was that OCX was “the most complex software ever created for space,” but other players in the sector have shown that achieving these types of technical milestones is possible. SpaceX has demonstrated this with technical “miracles” like its reusable Falcon 9 or with the development of Starship, for example, so those arguments are falling on deaf ears now. Waiting for a better GPS. These problems also affect us end users, who will not be able to enjoy the L5 signals for now. This much more robust frequency will significantly improve accuracy in urban centers with many tall buildings. The irony is tragic: we cannot use extraordinary space infrastructure because the base stations cannot cope with it. While waiting for the problems to be resolved, the learning is clear: the software cannot be a monster that takes 16 years to build In Xataka The GPS in the Baltic has been experiencing interference for months and the culprit is becoming increasingly clear: Russia And while as always, China. While the US crashes against its project to renew the GPS constellation, China has once again managed to “become independent” from Western technology. Your satellite navigation system Beidouit does not replace GPS, true, but It already complements it in 140 countries. Once again China’s long-term view has its obvious result: it has taken 20 years in deploying its constellation, but they already surpass the GPS system in metrics such as signal availability or integrated messaging services. Europe, by the way, also has its own alternative. In Xataka |GPS “dead zones” are spreading around the world: jammers are to blame for confusing drones (function() { window._JS_MODULES = window._JS_MODULES || {}; var headElement = document.getElementsByTagName(‘head’)(0); if (_JS_MODULES.instagram) { var instagramScript = document.createElement(‘script’); instagramScript.src=”https://platform.instagram.com/en_US/embeds.js”; instagramScript.async = true; instagramScript.defer = true; headElement.appendChild(instagramScript); – The news The US has invested 16 years and 8 billion dollars in renewing the software of its GPS network. Result: a failure of epic proportions was originally published in Xataka by Javier Pastor .

Jensen Huang believes he has found the perfect new bonus for software engineers. Not Stocks: AI Tokens

by

The CEO of Nvidia has been converting the AI tokens at the center of all their public conversations. Jensen Huang’s latest idea links these tokens to the efficiency of engineers and how the best engineers in the world are recruited: in addition to a generous salary, offer them an amount equivalent to half their annual salary in AI tokens as part of the hiring package.​ Huang verbalized his proposal during the inaugural speech of the GTC 2026 conferenceNVIDIA’s largest annual event for developers. In a later interview, the Nvidia CEO detailed that engineers would earn “a few hundred thousand dollars a year as a base salary,” and the intention would be to give them “probably half of that, also, in tokens, so they can multiply your productivity times ten”.​ What Huang proposes already has a name: Tokenmaxxing. In one podcast appearance ‘All-InHuang said he would be on “high alert” if an engineer earning $500,000 didn’t spend at least $250,000 a year on tokens. “If that person said (that he has used tokens worth) $5,000, I would go completely crazy,” Huang stated. When asked if NVIDIA planned to spend $2 billion on tokens for its engineering team, as proposed, Huang responded: “We’re trying.”​ As and how they counted in The New York Timesthat has generated a phenomenon called “Tokenmaxxing“, with which engineers brag about the number of tokens they consume to try to improve the perception of their productivity: the more tokens you consume, the more productive you are. Tokens as bonuses are a trend in Silicon Valley. The CEO of NVIDIA is not the only one who thinks this way, and the use of tokens as an extra work benefit it’s soaking among investors in the sector. Tomasz Tunguz of Theory Ventures stated to Business Insider that “companies are incorporating AI inference as a fourth component of engineer compensation: salary, bonus, stock and tokens.” The interest of whoever sells the chips. The NVIDIA CEO encouraging everyone to spend more on tokens is not disinterested advice. Gergely Orosz, analyst at software engineeringhe pointed it out bluntly in a publication from AND he added an analogy that sums it up accurately: “It’s almost as if the CEO of Apple said, ‘If someone who makes $500,000 a year doesn’t spend at least $50,000 a year on in-app purchases on iOS, I’d be deeply alarmed.’ And yes, you would be, because that would reduce the revenue you generate.” Huang is the head of the company that manufactures the chips for AI on which most of the world’s artificial intelligence runs. Huang himself made it clear to his investors: “Without computing, there is no way to generate tokens. Without tokens, there is no way to grow revenue,” he declared, describing his data centers as “token factories” whose demand will only grow as AI agents proliferate.​ Do not confuse value with price. However, Huang has incurred a bias when arguing his idea: confusing value with price. Orosz formulated it clearly in a message in X : “The advice that engineers should use tools that make them more productive IS correct… except that the cost of tools should NOT be what we focus on. Some of the most useful tools are very cheap. Of course, vendors will focus on selling the most expensive and most profitable tools.” Productivity is not measured in tokens spent, but in results achieved. The right question for companies should not be whether their employees use more AI, but whether increased use of AI is rewarded. with greater productivity. In Xataka | Customers demand that a human solve their problem. The surprising thing is that if humans serve them they think they are an AI Image | NVIDIA, Unsplash (Arif Riyanto)

The most profitable action of the AI ​​revolution in Spain is not a software company. It is a construction company

by

We know Florentino Pérez ample by hire galactics and for his business successes, but a priori we would not easily relate him to the rise of AI. And by not doing so we would make a serious mistake, because the manager managed to see before anyone else that this was a huge opportunity… and he is taking advantage of it almost without us realizing it. what has happened. ACS is a construction company that doesn’t seem particularly fascinating. You lay bricks, asphalt and cement, but in 2025 the data tells a fascinating story. The company obtained a net profit of 950 million euros, 15% more than the previous year, and the engine of that growth was its American subsidiary, Turnerwhose contribution to the group’s results grew by 66.6% to 549 million euros. Turner doesn’t build flats or highways. Build data centers. And therein lies the crux of the matter. AI needs big construction companies. The transformation has not happened all at once. ACS has been betting on this niche for years with a simple but powerful thesis: AI requires enormous amounts of hardware, and that hardware needs equally huge buildings with cooling, energy and security. And ACS is dedicated to precisely that: to build large buildings. In Xataka Amazon is building an empire in Aragon: it has just paid 1.5 million to expand the electrical network to its fifth data center Florentino triumphs in the US. Turner arrived earlier and stronger. In 2025, ACS won several large-scale data center contracts, including the construction of a 902-megawatt center in Wisconsin as part of the Stargate program, and a stake in the $10 billion, one-megawatt Meta campus in Indiana. Those are conventional projects. They are cities whose inhabitants are servants for this new era of AI. Go for it all. As they point out in five daysdata centers generated more than 9 billion euros in sales during 2025, and ACS has already delivered more than 9 GW of capacity all over the world. That figure is extraordinary, especially considering that in all of Spain the installed capacity barely reaches 7 GW. The Spanish company that talks the least about AI has been silently one of its great beneficiaries for years. Very much in the style of Florentino Pérez, who usually maintains a relatively low profile and succeeds without making too much noise. Stocks on the rise. The market took a while to see it, but it has reacted forcefully. ACS shares have soared 115% in the last twelve months. Today they are close to 110 euros and mark historical highs while the construction sector advances (“only”) 20%. Group sales they reached 49,848 million euros, with the US and Canada contributing 63% of the total. ACS is in practice more of a North American technological infrastructure company than a Spanish construction company. It is listed on the Ibex and is chaired by one of the great football personalities, yes, but its current driving force is not here, but in the US and in the AI ​​fever. Build and Own. ACS is not limited to executing other people’s contracts: it also wants to be the owner of what it builds. In January 2026, the company completed an alliance with Global Infrastructure Partners, BlackRock subsidiaryto create a 50/50 joint venture to develop a global data center platform with an initial capacity of 1.7 GW. Already before had bought Dornanan Irish engineering company specialized in this type of infrastructure, for 436 million euros. ACS doesn’t just want to build AI data centers: it wants to own a piece of that infrastructure. The dollar as a great risk. One of the big problems with this project is the US currency. With more than 60% of its income in North America, each fall of the dollar against the euro is a setback for the Spanish multinational. The devaluation of the dollar is already greater than 10% after the last twelve months, and that has prevented Turner’s growth from being even greater. According to Renta 4 analysts, the “currency effect” subtracted more than five percentage points from the growth of net profit. And investors warn. Analysts themselves consider that the AI ​​market has already discounted a good part of future growth. At Bloomberg, the consensus is to maintain the stock with an average target price of 88 euros, which would imply a fall of 20% compared to current levels. This is what usually happens with good economic stories: when everyone knows them, they are no longer an opportunity. But at ACS they are optimistic. Although experts are cautious, at ACS they expect that spending on infrastructure quadruples from now to 2034. In fact, they expect that the benefits of 2026 will go even further than those of 2025 and exceed 1,000 million euros. If it achieves this, Florentino’s company will have completed one of the quietest and most profitable industrial transformations in the recent history of our country. {“videoId”:”x86aas4″,”autoplay”:false,”title”:”60% of the INTERNET passes through HERE: This is the LARGEST Data Processing Center in SPAIN”, “tag”:””, “duration”:”266″} Turner is ahead. According to Data Center MagazineTurner accumulated a backlog – a portfolio of confirmed orders – of $39 billion as of August 2025. It is the dominant construction company in this segment globally, although of course it has direct competitors such as DPR Construction, Holder, Skanska or AECOM. However, none have achieved the same concentration of contracts with the hyperscalers (Meta, Amazon and Microsoft). Turner has been building its reputation as a builder of this type of facility for more than a decade, and it is very difficult to replicate that advantage quickly. The irony of ACS and Spain. There is a geographical paradox in this success story: Spain and Europe have years debating on digital sovereignty, technological dependence and the need to build own infrastructure for not to be left out of the AI ​​revolution. While this debate is taking place, the Spanish company that is most building this infrastructure is doing so almost exclusively outside of Spain. As … Read more

Mobile software has become so complex that Samsung has had to add an AI to explain it to you

by

There’s a feature in One UI 8.5 that says more than it seems. It’s called, in Samsung’s internal nomenclature, ‘voice settings assistant’: you ask the phone why the screen doesn’t turn off, and it explains which setting is causing it. You ask it why the volume goes down on its own, and it tells you where the setting is that controls it. In it briefing of the Galaxy S26 They mentioned it almost in passing, as a nice detail among bigger and more important news such as the 3 AIs in 1, or the spectacular screen and its privacy mode. But this also deserves some attention. For years, learning to use a cell phone was part of the deal. You scanned the menus, memorized where things were, got used to their quirks, and cursed when you changed brands and couldn’t find anything. The instruction manual disappeared many years ago because it was assumed that phones were already intuitive enough not to need it. And for a while, they have been. The problem is that phones have not stopped growing. Every generation, and I’m looking at both iOS and Android, adds settings, modes, features, and layers of customization. One UI 8.5 brings, in the AI ​​section alone, more than a dozen new functions. The Christmas tree effect: We accumulate things without getting rid of the previous ones and we end up with an unruly mammoth. The operating system of a modern mobile phone has thousands of options spread out in menus that are sometimes where you would expect and sometimes not. And when something behaves unexpectedly, finding the reason can take several minutes of searching or a simple query to Google. Or ChatGPT. Samsung has decided that the solution to this complexity is not to simplify, but add a layer to help you navigate it. The mobile phone no longer expects you to understand it: it explains how it works if you ask it. It is a pragmatic move. Manufacturers have been in a race for years to add functions that justify the annual update, and retracing that path would mean cutting features that some users do use. So the solution is not to remove, but to translate. An AI that acts as a guide within the device itself. Google already has similar functions in pure Android and the Siri that the prophets promised maybe one day it will come. What Samsung does with One UI 8.5 is go a step further: it not only takes you to the setting, but explains why that setting is affecting the behavior that misses you. It’s the difference between giving you directions and explaining the map to you. The question that remains is how far this goes. If the phone needs an AI to explain itself, The next logical step is for that AI to start making decisions for you.: not only explain why the screen does not turn off, but also turn it off when it detects that you are not using it. Some of the agentic upgrades that Samsung has presented in the S26 are already going in that direction. The cell phone that asks you what you want to do and the cell phone that deduces what you want to do are closer than they seem. In Xataka | Samsung Galaxy S26 Ultra, S26+ and S26, first impressions: a broken heart in an unprecedented commitment to AI Featured image | Xataka

There are those who claim that AI is going to kill software. Most likely, just the opposite will happen.

by

It was 1993 and a young man named Marc Andreessenstill with his hair intact and a lot of ambition, set out to create a web browser with a colleague who worked with him at the National Center for Supercomputing Applications (NCSA) at the University of Illinois. They called it Mosaic. That browser allowed you to explore the newborn World Wide Web with the click of a mouse, something amazing because to date the browsers that had been developed were in text mode and were used with the keyboard. Suddenly the web could include images and even multimedia content. A little later Andreessen met Jim Clark, founder of the legendary Silicon Graphics, and he encouraged him to embark on an adventure with his web browser, so together they decided to set up their own startup and that led to the creation of one of the mythical browsers in history: Netscape. Marc Andreessen. Source: Wikimedia. That made Andreessen a multimillionaire, and from 2005 his interest changed. I no longer wanted to start a business, but rather help others start a business. He ended up founding the venture capital firm Andreessen Howoritz and became richer and richer (while losing more and more hair). His successes and bets in the technology industry are notable, but also he left some famous phrases. The most notable is probably the one delivered in 2011 when saying “Software is eating the world.” His argument was compelling: the companies that were growing the most were software or had software as one of their key pillars. He was not wrong—today these companies are absolute technological giants—and that quote became one of those seemingly immutable laws. And then AI arrived. Is AI eating software? The appearance in November 2022 of ChatGPT caused an extraordinary impact, although it had been clear for a year and a half that something was changing in the software world. In July 2021 we talked about GitHub Copilot when the conversation around generative AI was still awakening. That project allowed machines to program for us. This concept has become over the years the clearest example that AI can change things– Developers have embraced this tool like no other industry, but they know that they cannot trust her 100%. Still, we are living in an exciting time for software. One in which the rise of vibe coding is absolute. Andrej Karphaty I thought about it these days and explained that when coined the term A year ago maybe he was wrong with that way of calling it. Now he proposed changing it to “agent engineering” to reflect the type of tool it has ended up being. Be that as it may, the vibe coding/agent engineering has sparked a fever for software development. In many ways it has democratized it and turned us all into potential developers. I am experimenting myself with Open Source tools that I am modifying to my liking, and others are doing exactly the same in this era of “custom micro-applications”. But in recent days we have also experienced a disturbing phenomenon. The threat of the “SaaSpocalypse” The generative AI models and AI agents that have appeared in recent months have ended up having an extraordinary impact on the software world. In fact we are not referring to vibe coding as suchmore aimed at occasional programmers or users without knowledge who are encouraged to create their own apps. We are referring to what has happened to the large software companies that for years have controlled the market with the SaaS (Software as a Service) philosophy. This model has made it possible to convert, for example, Photoshop or Office no longer into software that was sold in boxes and you installed on your PC, but into applications that run in the cloud and that you can use from a browser. Applications are no longer applications, they are services. And you don’t pay for them by buying them at once: you subscribe to them. But AI appears to threaten that model. Last week, software companies lost a total of $300 billion on the stock market overnight. The shares of MongoDB, Salesforce, Shopify or Atlassian lost between 15 and 20% in value in a few hours, and talk of the “SaaS apocalypse” began (“SaaSpocalypse“). Source: Perplexity These falls are obvious if you take a look at Google Finance or any monitoring platform for these companies. Or if you ask Perplexity like I did, which creates a nice (and worrying) graph about some of the companies consulted: the collapse in the last month is really terrible, although it seems that the trend seems to have stopped. Be that as it may, this “SaaS apocalypse”, whether it exists or not, raises a question that is precisely in line with what Andreessen said. If software ended up eating the world, Will AI eat software? Will it kill him? Software is not going to die. Just the opposite What is happening at the corporate level with these falls has of course to do with AI, but also with the model and philosophy of these companies themselves. Those SaaS platforms that dominate the world They have not stopped abusing their dominant position for years with aggressive price increases and rigid contracts. We have seen it with companies like Salesforce, whose customers have seen prices rise 35% in the last two years, or the mind-blowing case of Broadcom, whose customers in Europe were facing price increases of 1,500% in your VMware virtualization software licenses. This has created an ideal breeding ground for clients of these and many other companies with SaaS platforms to look for alternatives, and also look for them in AI. Artificial intelligence is not only offering efficiency, but is giving many customers that “key to the cell” that allows them to escape from their suppliers, who treated them like hostages. In fact, the current correction in stock market valuations can also be understood as a post-bubble hangover from 2021, when the pandemic boosted all these companies. We saw it with … Read more

Until now “software was eating the world.” Now AI is eating software

by

For years we repeated an idea that seemed indisputable: “software was eating the world.” It was the most direct way to explain why almost any sector ended up depending on an app, a platform or a cloud service. But something is beginning to change in a silent and, at the same time, tremendously ambitious way.: the artificial intelligence revolution is not only transforming entire industries, it is also putting pressure on the software industry from within. The question that begins to arise is delicate and fascinating at the same time: if AI can build custom tools in a matter of moments, what is the point of continuing to pay for rigid and standardized software that works, yes, but that often forces it to work as the platform dictates. This is the point at which the debate becomes really serious: it is not about incremental improvement, but about questioning the current model as the standard for enterprise software. The logic is aggressive, at least on paper. So we could be looking at a potentially massive change. And yes, “potentially” is the key word: there are reasons to think that this can happen, and equally strong reasons to believe that it can happen with very real limits. Software in times of artificial intelligence This may all revolve around a very earthly question: what are you paying for when you pay for software. Until now, the price included the construction of the tool, its evolution, and the cost of making it generic enough to sell to thousands of companies. If the AI ​​compresses that part and allows generate code fast and cheapthe value migrates to other places: flow design, real integration with business systems, measurable results. Bret Taylorfounder and CEO of Sierra and part of the board of OpenAI, insists that the focus must be on the value that the customer receivesnot in technology for technology’s sake. Until now, for most companies, the map was quite recognizable: either you bought a pre-packaged tool and assumed its rules, or you commissioned a custom development, usually slower and more expensive, but more tailored to what you needed. What AI introduces is an alternative that, on paper, breaks the balance: instead of choosing a piece of software, it would be enough to explain the problem and let an agent build a custom system, deploy it and adjust it as processes change. Bret Taylor describes it from Sierra’s experience with customer service agents: “Our hypothesis is that, if we move forward five years, the vast majority of digital interactions will be through an agent.” If that is true, the dominant interface of many companies would no longer be a traditional platform. Most importantly, this conversation no longer happens only at conferences or investor presentations. There are practical signs that the paradigm is, at the very least, emerging: the so-called “vibe coding” has become a reality for many non-developer users, capable of setting up a website or tools describing what they want with text. Platforms like the European Lovable They have pushed that idea to the general public: fewer technical barriersmore rapid iteration, less “project” and more trial and error. This does not mean that a company is going to replace its ERP by a system generated on the fly, but it does help to understand why the market and the industry are beginning to take the possibility seriously. And this is where enthusiasm often clashes with real enterprise. Corporate software does not live in isolation: it is attached to databases, legacy systems, identities, permissions, audits and integrations that have been working in a specific way for years. Added to this is the most delicate aspect: regulatory compliance, security and internal responsibilities, which in regulated sectors dictate what can be done and what cannot be done. Even if an agent can generate a functional system, it remains to be resolved who maintains it, who supports it, who ensures that it does not break over time, and who responds when something fails. In this area, “customized and fast” software still has many questions ahead. If all this still seems too abstract, Bloomberg provides a fairly clear thermometer: The market is already reacting as if the threat were real, although we still do not know how far it will go. The media explains that the launch of Claude Cowork on the part of Anthropic reactivated the fear of a disruption that puts pressure on traditional software. According to that text, a set of SaaS values ​​followed by Morgan Stanley as an indicator of the sector has fallen 15% so far in 2026 after falling 11% in 2025, the worst start since 2022. In addition to all this, some cited analysts suggest that right now there are no reasons to have shares of software companies in portfolio. Images | Hack Capital | Anthropic In Xataka | Meta was the big loser of the AI ​​race in 2025. She was actually preparing her big move In Xataka | AI has already destroyed the world of programmers as we knew it. Now it’s the turn of the translators

Microsoft no longer sells software: it sells inevitability

by

OpenAI is no longer an entity with hybrid control and is now a fully fledged company. That is, for profit. Microsoft, which had special rights and a seat on its board, give up that position in exchange for something more stable: Guaranteed and perpetual access to OpenAI models (current and future). Freedom to create your own foundational models without restrictions. Gain independence without losing technology. Why is it important. This does not make Microsoft the owner of OpenAI, but rather the platform that turns its AI into a mass product. OpenAI can continue investigating, but Microsoft remains the one who controls access to users and companies. Distribution defines power today, even above invention. The general overview. Microsoft has been transforming its business from selling licenses to selling continuous dependency for more than a decade: Office 365 eliminated or relegated the option to purchase the software only once. Windows 10 introduced mandatory updates that turned the operating system into permanent service. Azure has tied enterprise infrastructure to its cloud. The pattern is consistent: turning tools into platforms, products into subscriptions, and options into inevitabilities. The agreement with OpenAI is not an exception, it is the culmination. In detail. Microsoft maintains something that no other actor has: Direct integration of Copilot in Office, Teams, Outlook and Windows. Large-scale business contracts that turn AI into the structural cost of digital work. Control over the point of entry: the place where millions of people work every day. The new agreement ensures that OpenAI cannot turn off the tap, and that Microsoft can expand or replace its models without depending on third parties. The strategic background. Until now, Microsoft could not develop its own AGI. Now yes. This allows you two parallel routes: Use OpenAI models in your ecosystem. Develop your own (or integrate with others) if OpenAI gets sidetracked or delayed. Gain technological freedom and commercial stability. But above all, you gain something more valuable: the certainty that AI will not be optional in your software. Between the lines. The move consolidates Microsoft as the main consumer channel for AI at work. Not by contract, but by market position. Millions of users already pay for Copilot without expressly choosing it. Companies assume it as part of the normal cost of productivity. There is no real alternative: if you work in Word, you use Copilot. If you manage emails in Outlook, you use Copilot. If you coordinate teams in Teams, you use Copilot. Yes, but. This is not the traditional technological domain. Microsoft doesn’t need to have the best AI. You just need to have the most integrated one. OpenAI can be brighter, Google can be faster, Meta can be more open (or not so open). It doesn’t matter, because none of them are inside the software where the work is done. AI is no longer an add-on. It becomes invisible infrastructure. The contrast. Other technological giants continue to bet on the excellence of the model: Everyone competes to have the best technology, but Microsoft competes for something else: to be the place where that technology is used, regardless of who created it. In summary: OpenAI is freed to grow as a company. Microsoft makes sure that no matter what happens, AI runs through its software. The rest of the industry competes to invent. Microsoft has won by distributing. Does not sell AI. Sell ​​inevitability. In Xataka | AI works better if you are edge Featured image | Microsoft

Log In

Forgot password?

Forgot password?

Enter your account data and we will send you a link to reset your password.

Your password reset link appears to be invalid or expired.

Log in

Privacy Policy

Add to Collection

No Collections

Here you'll find all collections you've created before.