programmers

We believed that human programmers would end up being code reviewers. Anthropic just killed that

by

The rise of the Generative AI The world of software development seemed to follow a clear script: models would write the code and humans would review it. It was the new balance. Well, Anthropic just killed him. The problem of programming with AI. What we know today as vibe codingthis practice of giving instructions in natural language to an AI so that it generates code at full speed, has skyrocketed software production in companies. Anthropic affirms that the amount of code generated by each of its own engineers has grown by 200% in the last year. And now there’s a problem: there’s so much new code that reviewing it has become the bottleneck of the process. Human developers can’t cope. Many pull requests (change proposals that must be reviewed before integrating new code) are skimmed or not read very carefully at all. What Anthropic has done. The company Code Review has been releaseda tool integrated into Claude Code that, instead of waiting for a human to review the code, deploys a team of AI agents to do it automatically every time a pull request is opened. This new system is now available in preview phase for Team and Enterprise plan customers. Cat Wu, Product Manager at Anthropic, explained told TechCrunch that the question they constantly received from their clients’ technical managers was always the same: “Now that Claude Code is generating a ton of pull requests, how do I make sure they are reviewed efficiently?” How it works inside. AI agents work in parallel autonomously the moment a pull request is opened, examining the code from different perspectives. An end agent then aggregates and prioritizes the issues it has found, removing duplicates and sorting them by severity. The result reaches the developer through a featured comment, accompanied by more online comments about specific bugs. The focus, according to Anthropicis in logical errors, not in matters of style, something designed on purpose so that the feedback does not generate too much noise. Issues are labeled by color depending on how important they are: red for critical, yellow for attention, and purple for pre-existing code. Numbers. The company has been using Code Review internally for months before launching it to the market. According to what they saybefore implementing it, only 16% of their pull requests received meaningful review comments. With the tool, that percentage rises to 54%. In large pull requests (more than 1,000 modified lines) 84% returned results, with an average of 7.5 problems detected. And less than 1% of those results are flagged as incorrect by the engineers themselves. In one of the cases documented by the company, they spoke of a single line change that seemed routine. However, Code Review marked it as critical, as it apparently could have broken the entire service’s authentication. The bug was fixed before integration. Furthermore, according to the company, the engineer later acknowledged that he would not have caught it alone. ANDhe new role of the programmer. The narrative that had spread in the last two years was that developers would evolve towards a profile closer to that of a reviewer or supervisor of code generated by AI. Now that transition is also being automated, at least in part. Anthropic does not eliminate the human from the equation (in fact the tool does not approve pull requests), but it does compress the review work that was supposed to be the last bastion. It seems that now the human goes from reviewer to final arbiter. Price. It is not a cheap tool. Each revision has a cost based on token consumption. Anthropic esteem The average price per review is between $15 and $25, depending on the complexity of the code. It is a cost that the company justifies in the context of large technology companies where errors that escape review have a much higher price. Cover image | Compagnons In Xataka | Software companies sank on the stock market for a simple reason: investors are panicking about AI

Claude Code is being the big favorite among programmers. So much so that he already signs 4% of everything that is uploaded to GitHub

by

It is worth taking a look at how generative AI It is transforming the daily lives of many programmers. And little by little these tools are conquering the environments of millions of developers. The achievement in this aspect is for Claude CodeAnthropic tool, which already represents 4% of all public commits uploaded to GitHub, according to a report by SemiAnalysis. The media says that, if it maintains its current pace of adoption, it is very possible that it will reach 20% of all daily contributions before the end of 2026. Although there are nuances that should be highlighted. Why is it important. Claude Code is slowly gaining the reputation of being the favorite tool for programming with AI. The tool works radically differently than traditional code wizards. It is not a chatbot integrated into an editor like Cursorbut rather a terminal tool that reads entire code bases, schedules multi-step tasks, and executes them with full access to the developer’s computer. You can start from spreadsheets, entire repositories, or web links, understand context, verify details, and complete complex objectives iteratively. The interesting thing is that, by default, Claude Code includes a co-authorship note if the user has used this tool in their program and uploads it to Github. But the user can also decide not to include that signature if modify the parameters by Claude Code, so that 4% could remain small. In March of last year, a month after its launch in private beta, Claude Code already had the co-authorship of about 15,000 Github commits in a period of 48 hours. Things have ended up escalating quickly. Opinions. The newsletter stands out the comments of some industry professionals regarding the vibe codding. Andrej Karpathy, one of the first to coin the term vibe codding, recognized in a post that he is “starting to lose the ability to write code manually.” Ryan Dahl, creator of Node.js, counted directly that “the era of humans writing code is over.” Boris Cherny, creator of Claude Code, assures that “practically 100% of our code is written by Claude Code + Opus 4.5“. Even Linus Torvalds, creator of Linux, has fooled around with vibe codding for some of his personal projects. It should be noted that, despite all the benefits of Claude Code, it is not perfect. Already we pointed out some time ago the words of Kelsey PiperAmerican journalist for The Argument, who explained that 99% of the time using Claude Code is like having a magical, tireless genie, but 1% of the time it’s like yelling at a pet for peeing on the couch. He can and does make mistakes. It also gets stuck. Hence, the expertise of the person who uses it also plays a very important role. Beyond programming. There is an increasingly latent threat with the use of AI tools (well there are a few that accumulate already). And according to account SemiAnalysis, any information work that follows the READ-THINK-WRITE-CHECK pattern can be automated with this technology. The report mentions sectors such as financial services, legal, consulting and data analysis, which add up to billions of workers globally. Anthropic has already taken the next step with coworkreleased a few weeks ago, which is basically Claude Code applied to general office work. According to the company itself, Cowork was developed by four engineers in ten days, mostly with code generated by Claude Code himself. The tool can create spreadsheets from receipts, organize files by content, write reports from scattered notes… And all with access to your computer. The big consultancies and AI. In December, Accenture signed an agreement to train 30,000 professionals on Claude, the largest deployment of Claude Code to date. OpenAI, for its part, Frontier has launched focused on business adoption so as not to lose steam in the field of corporate use of AI, a business that can end up being very lucrative for startups. Cover image | Anthropic and Mohammad Rahmani In Xataka | Programming is the new board of AI. OpenAI and Anthropic have made it clear with GPT-5.3-Codex and Claude Opus 4.6

AI has already destroyed the world of programmers as we knew it. Now it’s the turn of the translators

by

On November 8, 1519, an extraordinary meeting took place: Hernán Cortés met with Emperor Moctezuma II. Of course, neither one nor the other understood anything of what their interlocutor was saying: Hernán Cortés spoke Spanish and Moctezuma spoke Nahuatl, but that problem was solved thanks to two chain translators: Malinche translated from Nahuatl to Mayan, and Jerónimo de Aguilar went from Mayan to Spanish, and vice versa. History is full of legendary translations like that one, and in all of them, human beings depended on human translators to understand the other party. That has been changing with various technologies, but the one that is really about to change everything is AI. With AI we have found (and translated) In fact, translation technology has run parallel to technological evolution itself. From the translation based on rules of the second half of the 20th century we moved in the 90s to the automatic statistical translations which, for example, ended up using Google Translate. These systems looked for the “most likely” translation, not the “most correct” one. These statistical models improved with the phrase-based translationbut The final leap was made by DeepLwhich appeared in 2017 to change everything with the use of neural networks and neural machine translation. Google had also started to adopt that system in 2016, and it was clear what the path was. With the arrival of generative AI we have found ourselves with another potential leap in this field. There are, however, differences: these systems are based on large language models (LLM) that are then trained and tuned specifically for translationwhich a priori gives them an advantage when it comes to achieving more natural and versatile translations. The application of AI models to the field of translation seems to be following in the footsteps of what we have seen with programming. Developers have embraced this revolution and many of us have realized it thanks to the vibe coding that it is possible to program without knowing how to program. The same clearly occurs with these systems that enable us to know how to speak languages ​​that we don’t actually know how to speak. Machines do it for us, and they do it better and more immediately. The real-time translation is very fashionable and both Google and Meta—which has been warning for a long time— they are integrating it into their current or future glasses augmented reality. Apple, which does not usually launch things that are not mature, has just integrated it on your AirPods. The user experience may not perfect at the momentbut it is clear that this type of function is going to become more and more common, a commodity technological more. The transition And this transition that wants to turn access to quality translations into something “trivial” has been made evident these days with the launch of two platforms. The first, the ChatGPT Translatorwhich is surprising not because it is an obvious and simple use case for AI, but because it is a logical indiscriminate copy of the services that already work, Google Translate and DeepL. Being able to do the same with AI shows that that problem seems solved. The translation of Gemma 3 27B was already good. TranslateGemma’s is even better, even with smaller models and challenging language pairs. And if it didn’t seem like it enough, Google has just presented its new generative AI models specifically aimed at translation. It is about TranslateGemmaa family with versions 4B, 12B and 27B (the latter, logically, the most capable) that allow these tasks to be carried out locally, privately and without connection to the cloud. They support 55 language pairs and of course they are prepared for the most popular ones (English, Spanish, Chinese, French, Hindi), but their creators already indicate that they are training them with 500 additional language pairs for the future. We are therefore facing a moment in which learning a language will probably end up becoming something more vocational or aspirational than something that we really need on a daily basis. Human translators, like human programmers, will still have valuebut once again what is clear is that AI is going to make this type of capability more accessible than ever. In Xataka | Some of the emails you read may not say exactly what was written. A forgotten Gmail setting is to blame

More and more programmers depend on AI to program. And every time they trust her

by

Programmers love AI, but they don’t trust her too much. This is confirmed by a recent survey that Stack Overflow has done and in which 49,000 professional developers in their community have participated. That conclusion is as contradictory as logic, and points to a potential transformation of this sector. Each time they use it more. According to him Complete study84% of developers already use AI as part of their workflow, when last year that proportion was 76%. The proportion is in the line of a survey conducted in 2023 in the github community, although in that case of the 500 programmers surveyed, 92% confessed Use AI tools to program. But every time they trust less. The other prominent data of the survey is the one that indicates that programmers trust somewhat less in the Code generated by these AI tools. If last year the confidence in precise solutions was 40%, this year that confidence is only 29%. I spend the day correcting mistakes. The most important frustration of developers is that they are working with AI solutions that gives a rather correct, not completely correct answer. That implies that in the end developers must devote much more time to detect and correct those errorsand in fact 66% of them confess to investing longer to fix that “almost correct” code of AI. I trust more than human experts. When correcting errors there is another unique conclusion: in complicated code fragments, 75% of respondents claim that They would ask another human programmer (and not to another model of AI) when they do not trust the answer or the code generated by the machines. AI agents do not set so much. Although tools “Vibe Coding“As a cursor or Windsurf, they have positioned themselves as a very interesting option even for new programmers, that theoretical revolution is far from being a reality. Of course: they gain productivity, and 69% say they have seen said metric increased thanks to agents. Will I replace an AI? Programmers continue to see these tools more as a help and assistance than as a possible substitution. The majority (64%) do not see this technology as a threat to their work, but it is also true that this percentage was somewhat higher, of 68%: there is a small increase in that threat than For Jensen Huang, CEO of Nvidia, it is inevitable. Learning to program. This community also made it clear that it does not stagnate: 69% of them have invested time in learning new programming techniques or a new language. Here is another relevant fact: 44% have learned new things thanks to AI tools, when 37% did it last year. Work correcting the code of a machine. The survey seems to point to a future in which programmers end up programming less and less to become something like project chiefs or software engineers. His work will no longer be the one to chop code, but probably that of Check the code generated by these systems of artificial intelligence. Image | Sigmund In Xataka | The AI is opening the doors of a radical revolution on the Internet: that we can all create apps without knowing

Google’s strategy to be the perfect partner of writers and programmers

by

Artificial intelligence chatbots (AI) have come to lend us a hand when words do not come out or the code stuck. They can help write that speech that does not finish flowing or solving doubts in the middle of a programming line. But there is a problem: to move forward with Prompts, one after another, it can end up being an obstacle, especially when we want to focus on a concrete part of the project. So much Chatgpt as Claude They have faced this challenge With a function that acts like a canvas, allowing to work directly on specific parts of the content. Now is Google’s turn, which has just presented the canvas of Gemini. As in the other proposals, we find an interactive window aimed at helping us to improve our work in real time thanks to several commands and format tools. It all starts with a draft. Returning to the example we mentioned at the beginning, if you are thinking of writing a speech to honor someone special, you can start creating a rather basic draft and gradually work the text. We will not have to elaborate a very precise prompt to indicate to Gemini to modify this or that part. It will be enough to mark a paragraph and ask him to rewrite him with a new style. And in programming? If we are working in software development, Gemini Canvas also promise to be very useful. In addition to receiving the classic, it helps to chop code that the wizard can offer us, the “canvas” will allow us to generate and prevail HTML/React code. This will result in that we can have a visual prototype of our web applications, which should help us in programming tasks. But this does not end in the preview. Once we have the preview of our project we can ask Gemini to modify input fields or add action buttons more easily. The combination of Gemini + Canvas should allow us, according to Google, work with Python scripts, as well as creating games that we can try directly in the browser window and run simulations. When will Canvas available in Gemini? Google has just announced the function and is now available worldwide, including the European Union. Acting it is as simple as having a Gemini or Gemini Advanced account. Once inside, you just have to look at the bottom of the chatbot text box, where a new button next to Deep Research now appears. From there, you can directly access this tool. Images | Google In Xataka | The videos of AI have broken the Instagram and Tiktok algorithms. Welcome to the new “AI landscape”

Young programmers no longer know how to program. AI is now causing the same as the calculator caused half a century ago

by

Myron Aub was a boiler man from the egg head, but all that did not matter. I had discovered how to do something that humanity had forgotten: I knew how to multiply without the help of a computer. After demonstrating it, the members of the new Pentagon were amazed: someone was able to multiply with a paper and a ball! All that imagined Isaac Asimov in his short novel ‘The feeling of power‘, originally published in February 1958. History, in just 3,700 words, is simply prodigious – I encourage you to read it – and raise a future in which human beings do not know how to perform mathematical operations and depend on computers for computers all. When AUB performs reverse engineering of this calculation process, it creates an extraordinary situation. One that leads to an equally surprising conclusion … and sadly predictable. I read that story about 30 years ago and then I found it fascinating – I am an absolute fan of the ‘Foundation’ saga of Asimov, but I hated the Apple TV+ series for betraying that legacy. The prolific author transmitted a clear message: perhaps it is not so good idea to depend too much on the machines. Or maybe yes? The programmers who did not know how to program It is what is certainly happening in the world of programming, which is undoubtedly the segment most affected by the arrival of artificial intelligence. The AI ​​models have proven to be valuable attendees when programming, and just six months after the chatgpt launch nine out of ten professionals used AI to program. The de facto models have become increasingly capable in this type of task, and the popularity of Tools as cursor has shown that The conquest of “effortless programming” Tab-Tab-Tab is increasingly clear. Even those who did not know how to program or knew, but not in certain programming languages, they are now capable of creating surprising applications. AI is not perfect, of coursebut one thing is clear: it is getting better. That, of course, raises a risk: let’s forget how to program. It is just what a developer named Namanyay Goel denounces, which In his blog he explained How “New Junior developers are not able to program.” The AI ​​allows developers to deliver more code than ever, but these young programmers do not know why that code works or if there would be another way to do things. According to their experience, the new generation of programmers use chatgpt or co -pilot or Claude at all hours. That, he says, allows them to deliver more code than ever, but according to this developer, These young programmers do not know why that code works or if there would be another way better to do it. “We are sacrificing deep understanding (of the code) by fast patches, and although that makes us feel good now, we will pay later.” Goel highlighted how not much places like Stack Overflow were a much better source of information for programmers. They asked things, but when they got answers they used to learn why those answers were valid. That knowledge was there for free, and also in many cases veteran developers and experienced people became involuntary teachers for new generations. “The AI ​​gives you answers, but the knowledge you acquire is superficial. With Stackoverflow, you had to read multiple expert discussions to get a complete vision. It was slower, but you ended up understanding not only what worked, but why it worked.” Of course, not everyone thinks the same and a commentator in Slashdot He pointed out How “Stackoverflow has been a source of terrible programming tips and an overdependence of copying and hitting for much longer.” However, Essste developer believes that not yet everything is lost. AI can help you, without a doubt, but you can also continue learning with it. “When you give you an answer,” he advised, “Ask him for her. Ask why (he has given that solution). “He also recommends debating that code with your team of developers to discuss and be able to get new ideas, or perhaps do so with developers who go to platforms such as Reddit, Discord or Mastodon. The calculator teaches us the future The reflection of that developer is undoubtedly striking, but for many it is a useless debate. In Slashdot a user named Zak3056 He affirmed Take two decades interviewing Júnior developers. “A surprising percentage of them did not understand basic concepts (…). AI has nothing to do with this. It is the state in which education is, many schools are creating graduates that simply do not understand the field they have chosen “ Other comments coincided with him, and the situation reminds other industrial and technological revolutions in which a profession or discipline have ended up being completely conquered by machines. We continue learning to perform mathematical operations, but after school there are not many people who do operations at hand: It is much faster to use the calculator. The calculators may cause some workers who performed that task to be displaced at the beginning, but their impact in the long run has been greatly positive. The calculators may cause some workers who performed that task to be displaced at the beginning, but their impact in the long run has been greatly positive To begin with, there are still mathematicians, but they are dedicated to much more complex problems than the machines are not yet able to solve – of the moment either – and the calculators made them even more valuable: they could focus on those problems, and Not in the calculations that could also be wrong because we assume it, the margin of human error is there. In addition, the calculators productivity and efficiency improved significantlysaving all kinds of professionals those most repetitive tasks to focus on others that other non -substitutable capacities (at least, for the moment) by the machines. Will something like this happen with AI and programming? It is very likely. AI has … Read more

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.