<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Mario Bogarin on Medium]]></title>
        <description><![CDATA[Stories by Mario Bogarin on Medium]]></description>
        <link>https://medium.com/@mbogarin?source=rss-7922852b0fd5------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*hdYlgE5U8POnt9ob</url>
            <title>Stories by Mario Bogarin on Medium</title>
            <link>https://medium.com/@mbogarin?source=rss-7922852b0fd5------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 02:51:54 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@mbogarin/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Full-Stack PMs: Expanding Influence When AI Lowers Execution Costs]]></title>
            <link>https://medium.com/@mbogarin/full-stack-pms-expanding-influence-when-ai-lowers-execution-costs-e2cdef48f1cd?source=rss-7922852b0fd5------2</link>
            <guid isPermaLink="false">https://medium.com/p/e2cdef48f1cd</guid>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Mario Bogarin]]></dc:creator>
            <pubDate>Sun, 15 Feb 2026 20:32:08 GMT</pubDate>
            <atom:updated>2026-02-15T20:32:08.172Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XWeTiHGoXuMVbv4dIynWeg.png" /><figcaption>* AI Generated image</figcaption></figure><p>In a previous article, I argued that <a href="https://medium.com/@mbogarin/why-product-managers-need-to-ship-code-now-09176e368563?sk=340a74fd7611e4892c3631c23d98dedb">Product Managers need to ship code</a>. The core idea was simple: in a world where execution tools are becoming more accessible, PMs who remain purely coordinators risk becoming bottlenecks instead of accelerators.</p><p>But that argument was only the first layer.</p><p>Shipping code is not the end goal. It’s the beginning of a broader shift in how product teams operate, now that AI is fundamentally changing the cost and speed of execution.</p><p>What we’re starting to see is not just more technical PMs. We’re seeing the emergence of a new archetype: the Full-Stack PM.</p><h3>From Overlap to Compression</h3><p>For years, product organizations were structured around three clearly defined domains: Product, UX, and Engineering. Each had its area of expertise, and collaboration happened in the overlap (<em>Of course, there are other disciplines involved, IT, DevOps or SREs, but let’s keep them apart for the shake of simplicity)</em>.</p><ul><li>PMs defined the “what” and the “why.”</li><li>Designers shaped the experience.</li><li>Engineers built and maintained the system.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*G_yeJjPz2m_BTh5jQF4i_w.png" /></figure><p>This model made sense when building software was expensive and iteration cycles were long. The cost of making changes justified formal handoffs and rigid role boundaries.</p><p>But AI compresses execution.</p><p>Today, changing a copy, creating a landing page, building a prototype, or validating a simple workflow may take minutes instead of days. The implementation cost of small ideas has collapsed. What hasn’t fully collapsed yet is the coordination cost.</p><p>And that mismatch is where friction now lives.</p><h3>AI as the New Center of Gravity</h3><p>If you were to redraw the classic Venn diagram of PM, UX, and Engineering, AI would now sit in the middle. It acts as a shared capability layer that reduces dependency for lightweight execution.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yEfE4aH5T7eMOSRackJvhA.png" /></figure><p>This doesn’t eliminate roles. It doesn’t make engineers or designers less important. What it does is lower the threshold for execution. <strong>The old liturgy of prioritization is, somehow, out of the game.</strong></p><ul><li>A PM can now prototype an idea directly.</li><li>They can generate and test copy variations.</li><li>They can build small internal tools.</li><li>They can validate assumptions before mobilizing engineering.</li></ul><p>The result is not role replacement, it’s role expansion.</p><p>The Full-Stack PM is simply the professional who embraces that expansion.</p><h3>Expanding the Area of Influence</h3><p>One of the notes I wrote to myself while thinking about this was: “<em>Their area of influence expands.”</em></p><p>That’s the essence of the shift.</p><p>A Full-Stack PM is not trying to become a senior engineer or a product designer. Instead, they are increasing their execution surface area. They reduce the number of times they need to say, “I’ll open a ticket for that,” and increase the number of times they can say, “Let me test this quickly.”</p><p>This has two important consequences.</p><p>First, learning cycles accelerate. Small experiments no longer compete with roadmap priorities because they don’t always require engineering bandwidth.</p><p>Second, engineering time becomes more protected.</p><h3>What This Means for Developers</h3><p>There’s often an implicit fear that if PMs become more execution-capable, developers lose relevance. I think the opposite is true.</p><p>Developer time is extremely valuable. It should be invested in what only strong engineers can do well: architecture, data design, core domain logic, scalability, security, and long-term system integrity.</p><p>No high-performing engineering team should be overloaded with minor copy changes or lightweight UI adjustments if those can be handled safely elsewhere. The more operational noise we remove from engineers, the more space they have to operate closer to the business and to tackle high-leverage problems.</p><p>In this model, developers don’t shrink in importance. Their focus shifts upward. <strong>Developers also become full Stack PMs.</strong></p><h3>The Cultural Shift Behind the Technical One</h3><p>The deeper transformation isn’t technical. It’s cultural.</p><p>Traditional product development created subtle ownership walls. Even when collaboration was strong, there were implicit boundaries around “who does what.” Those boundaries made sense when skills were scarce and tooling was rigid.</p><p>AI blurs those edges.</p><p>When execution becomes easier, the real constraint is no longer implementation capacity. It’s judgment. It’s taste. It’s prioritization. It’s domain understanding.</p><p>The most valuable people in this new environment are not those who can produce code the fastest, but those who understand what is worth building and why.</p><p>That’s why the Full-Stack PM is not defined by coding ability alone. They are defined by context fluency combined with execution autonomy.</p><h3>From Shipping Code to Orchestrating Outcomes</h3><p>If the first step was “PMs should ship code,” the next step is broader.</p><p>The Full-Stack PM is not just someone who can implement features. They orchestrate outcomes across business, design and technology, using AI as a leverage layer.</p><p>They reduce friction between idea and validation. And they do this not by replacing anyone, but by absorbing low-friction execution into their own workflow.</p><p>In a world where AI continues to compress the cost of building, the competitive advantage shifts away from pure execution and toward clarity of thought and quality of decisions.</p><blockquote>Execution is becoming cheaper. Context is not.</blockquote><p>And the Full-Stack PM is, ultimately, a response to that reality.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e2cdef48f1cd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI-Generated Apps Are the New Excel Macros]]></title>
            <link>https://medium.com/@mbogarin/ai-generated-apps-are-the-new-excel-macros-8745a97a1606?source=rss-7922852b0fd5------2</link>
            <guid isPermaLink="false">https://medium.com/p/8745a97a1606</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[corporate-culture]]></category>
            <dc:creator><![CDATA[Mario Bogarin]]></dc:creator>
            <pubDate>Sun, 01 Feb 2026 08:39:02 GMT</pubDate>
            <atom:updated>2026-02-01T22:01:32.166Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="AI generated apps are the new excel macros" src="https://cdn-images-1.medium.com/max/800/1*mbIQFslGcsO7wnCbaUfa6g.png" /><figcaption>*Image created with AI</figcaption></figure><p>For years, there has been an uncomfortable truth inside almost every medium or large organization, one that everyone knows and very few document:</p><p>Critical processes don’t always live in the “official” systems. They live in Excel, in Office macros, local scripts.</p><p>Massive spreadsheets. Nested formulas no one fully understands. Macros written years ago. Files named something like FINAL_v7_REAL_FINAL.xlsx.<br>And yet… they work. Or more accurately: <em>they work because a specific person makes them work</em>.</p><p>I’ve seen this happening in so many areas: finance, operations, engineering, logistics, or HR. They are not software engineers. They don’t write tests or deploy pipelines. But they deeply understand the business and have enough courage to bend the tool until it matches reality. Actually, for these people, building their own macros and scripts was the only way to scale certain processes. Honestly, in many cases, those needs were not “sexy” or important enough for engineering teams to take care of them.</p><p>For a long time, Excel was the ceiling for that kind of creativity.</p><p>That ceiling is gone.</p><h3>From macros to apps</h3><p>We are now seeing something different: non-technical people building <strong>full applications</strong> with the help of AI.</p><p>Not toy demos. Real things:</p><ul><li>Forms with non-trivial logic</li><li>Approval workflows</li><li>Dashboards connected to live data</li><li>Automations replacing manual processes</li><li>Internal tools that people actually depend on</li></ul><p>Tools like Replit, Cursor, Vercel, low-code/no-code platforms, and large language models have <strong>dramatically expanded the reach of creative people inside organizations</strong>.</p><p>Before, that creativity ended in a macro. Now, it can end in an app.</p><p>In practice, AI becomes a new intermediate layer between business understanding and software.</p><h3>This isn’t new. The scale is.</h3><p>This phenomenon itself is not new.</p><p>There has always been shadow IT.<br>There have always been parallel solutions.<br>There has always been someone “fixing things on the side.”</p><p>What <em>is</em> new is:</p><ol><li><strong>Speed: </strong>an idea can become something functional in hours</li><li><strong>Perceived quality:</strong> it doesn’t look like a hack anymore; it looks like “real software”</li><li><strong>Accessibility</strong>: you don’t need to know how to code, just how to explain the problem</li><li><strong>Complexity: </strong>you can write in almost any programming language, any environment.</li></ol><p>Combine that with business pressure, overloaded IT teams, and large organizations and the outcome is predictable.</p><p>It will happen. In many companies, it’s already happening.</p><h3>The default reaction: ban it</h3><p>The first reaction from many organizations is almost automatic:</p><blockquote><em>“This is risky.”</em></blockquote><p>And they’re not wrong.</p><p>Once these apps start spreading, very real issues appear:</p><ul><li><strong>Security: </strong>hardcoded credentials, uncontrolled access, sensitive data exposure.</li><li><strong>Scalability:</strong> something built for five users ends up serving fifty</li><li><strong>Errors:</strong> no reviews, no validation, no monitoring</li><li><strong>People dependency:</strong> if the creator leaves, no one knows how it works</li></ul><p>From the outside, the conclusion seems obvious: stop it and let IT prioritize and build tools as we always did.</p><h3>If you ban it, it won’t disappear</h3><p>We’ve seen this dynamic many times before, and it’s rarely about defiance or rule-breaking.</p><p>People inside organizations are trying to accomplish concrete goals under real constraints. They use the tools they have access to, at the speed their context demands. When a problem needs solving, the question is not <em>“what is the ideal tool?”</em> but <em>“what can I use right now?”</em></p><p>Sometimes that tool is Excel. Sometimes is that colleague that knows a bit of python. Increasingly, it’s an AI assistant that can turn intent into working software.</p><p>When existing tools don’t fully match reality, people don’t stop working. They stretch what’s available. They combine systems. They create small solutions that bridge the gap between how the organization operates on paper and how it operates in practice.</p><p>And when it happens, it usually does so <strong>without visibility, without support, and without safety nets</strong>.</p><p><strong>Now, the biggest risk is not that people build these apps. </strong>The biggest risk is that they build them <em>in the shadows</em>.</p><h3>The wrong question</h3><p>Most discussions start with the wrong framing:</p><blockquote><em>“How do we prevent this?”</em></blockquote><p>That’s the wrong question.</p><p>The right one is:</p><blockquote><em>“How do we create the right framework so this can happen safely?”</em></blockquote><p>Because whether we like it or not, AI-generated apps are the modern equivalent of Excel macros.</p><p>And macros never went away. Somehow they became business-critical.</p><h3>Governance is not control</h3><p>When people hear “governance,” they often think of bureaucracy. In reality, good governance is about <strong>channeling energy, not suppressing it</strong>.</p><p>Some practical ideas:</p><ul><li><strong>Toolbox: </strong>provide a list of available AI tools. Make it easy for employees to use them.</li><li><strong>Safe zones:</strong> what kinds of apps can be built</li><li><strong>Clear boundaries:</strong> which data can be accessed, and which cannot</li><li><strong>Shared foundations: </strong>authentication, logging, data access already solved</li><li><strong>Lightweight reviews:</strong> not to block, but to catch obvious risks</li><li><strong>Promotion paths:</strong> how a useful tool becomes “official software” when it grows</li></ul><h3>The changing role of technical leadership</h3><p>There’s a deeper implication here, especially for engineering leaders.</p><p>For years, engineering’s role was to <em>build software</em>. Increasingly, it will also be about <strong>designing systems in which others build software</strong>.</p><p>That means:</p><ul><li>Thinking in platforms, not just products</li><li>Accepting that not all code will be written by your team</li><li>Designing guardrails instead of locked doors</li></ul><p>This isn’t a loss of control. It’s a smarter distribution of it.</p><p><strong>Extra ball:</strong> Now think in not only other folks building and using software but in Agents and how they interact with what engineers build. I wrote some ideas about it. <a href="https://medium.com/@mbogarin/what-is-an-ai-agent-and-why-it-will-change-how-you-build-software-150ee5291d61?sk=e02c8f19b29c78312a8b22adec30b04d">Read the article</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8745a97a1606" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Rise, Fall, and Return of the Software Generalist]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@mbogarin/the-rise-fall-and-return-of-the-software-generalist-1380c6ec3a4a?source=rss-7922852b0fd5------2"><img src="https://cdn-images-1.medium.com/max/1536/1*LbmpYcn-PD_U40FDsAPfYA.png" width="1536"></a></p><p class="medium-feed-snippet">In the 1980s, a single person often wrote the code, talked to users, and decided what to build. Today, it sometimes feels like we need six&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@mbogarin/the-rise-fall-and-return-of-the-software-generalist-1380c6ec3a4a?source=rss-7922852b0fd5------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@mbogarin/the-rise-fall-and-return-of-the-software-generalist-1380c6ec3a4a?source=rss-7922852b0fd5------2</link>
            <guid isPermaLink="false">https://medium.com/p/1380c6ec3a4a</guid>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[ai-tools]]></category>
            <category><![CDATA[product-development]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Mario Bogarin]]></dc:creator>
            <pubDate>Sat, 24 Jan 2026 22:19:11 GMT</pubDate>
            <atom:updated>2026-01-24T22:19:11.042Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Why Product Managers Need to Ship Code Now]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@mbogarin/why-product-managers-need-to-ship-code-now-09176e368563?source=rss-7922852b0fd5------2"><img src="https://cdn-images-1.medium.com/max/1024/1*DmXm8x7_y2SHuuM2e9VdsQ.png" width="1024"></a></p><p class="medium-feed-snippet">Not because they&#x2019;re bad at their jobs. Not because documentation, discovery, or strategy stopped mattering. But because the cost of not&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@mbogarin/why-product-managers-need-to-ship-code-now-09176e368563?source=rss-7922852b0fd5------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@mbogarin/why-product-managers-need-to-ship-code-now-09176e368563?source=rss-7922852b0fd5------2</link>
            <guid isPermaLink="false">https://medium.com/p/09176e368563</guid>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Mario Bogarin]]></dc:creator>
            <pubDate>Mon, 19 Jan 2026 20:53:24 GMT</pubDate>
            <atom:updated>2026-01-19T20:53:24.094Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Will AI Kill Software Development?]]></title>
            <link>https://medium.com/@mbogarin/will-ai-kill-software-development-54d7e0f831bd?source=rss-7922852b0fd5------2</link>
            <guid isPermaLink="false">https://medium.com/p/54d7e0f831bd</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Mario Bogarin]]></dc:creator>
            <pubDate>Sun, 13 Jul 2025 18:56:03 GMT</pubDate>
            <atom:updated>2025-07-13T18:56:03.352Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fvEePs4SsnI0Lm-zUIqGpQ.jpeg" /><figcaption><em>Image from TyliJura at Pixabay</em></figcaption></figure><p>Everyone’s talking about artificial intelligence. About AI agents, copilots, foundation models and the infinite possibilities this technology promises. In the world of software development, the conversation is vibrant. Are we entering a new era of productivity — or witnessing the beginning of the end for the programmer’s role?</p><p>In recent years, we’ve seen an accelerated evolution of development tools powered by AI. Some of these tools are already regular companions for many professionals in the field. Everyone knows GitHub Copilot, Cursor, or Windsurf. They’re powerful and useful tools, capable of significantly boosting the productivity of an individual developer — or even an entire team. (We’ll leave for another time the discussion about how to measure this productivity gain, which is far from trivial.)</p><p>At the same time, a whole new generation of no-code and low-code tools is emerging with force: platforms like Replit, Lovable, v0.dev, or Databutton allow people with no technical background to create, launch, and maintain functional applications. With increasingly accessible interfaces and the help of generative models, it’s easier than ever to turn an idea into working software — without writing a single line of code.</p><p>But this isn’t new. It’s happened before. And this is where a fascinating concept comes into play: <strong>Jevons’ Paradox</strong>.</p><h4>Jevons’ Paradox</h4><p>Jevons’ Paradox was proposed in the 19th century by economist William Stanley Jevons. His theory states that when a technology increases the efficiency of using a resource, it doesn’t necessarily reduce the consumption of that resource. In fact, the opposite often happens: as it becomes cheaper to use, demand increases — and total consumption goes up.</p><p>Jevons applied this to coal: new steam engine designs were more efficient and required less coal to produce the same energy. But the result was that more factories and businesses started using these machines, and overall coal demand skyrocketed.</p><p>This pattern has repeated with many other resources. A classic example is gasoline: despite decades of improvements in engine efficiency, the number of vehicles and total fuel consumption have continued to grow.</p><h4>So what about software?</h4><p>Something very similar is happening (and will continue to happen) in the world of software. In recent years, the pace of software creation has exploded. Just look at the data GitHub publishes each year: the number of repositories, contributions, active users, and launched projects has grown steadily and significantly.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RFaETl11Fl8h9VaZfP-2lw.png" /><figcaption>** Reports from github: <a href="https://octoverse.github.com/2022/developer-community#:~:text=%2A%2094M%20developers%20with%2027,to%20all%20projects%20on%20GitHub">2022</a> and <a href="https://github.blog/news-insights/octoverse/octoverse-2024/#:~:text=%27108M%20new%20repositories%20in%202024,gradient%20colors%20and%20glowing%20text">2024</a></figcaption></figure><p>And that’s just one part of the picture. There are more apps, more digital services, more platforms, more automation in businesses of all sizes and sectors. The trend is clear: there’s a <em>huge</em> amount of software still to be built.</p><p>So if demand for software is so high — why aren’t we producing more of it?</p><p>Two main reasons:</p><ul><li><strong>Software is expensive</strong>: It requires skilled professionals, long development times, testing phases, maintenance, updates, support, and more.</li><li><strong>Change is slow</strong>: Even when the software is ready, rolling it out in an organization requires training, adapting workflows, migrating data, and gaining user adoption.</li></ul><h4>What to Expect from AI</h4><p>These two barriers limit how fast companies and organizations can adopt new technology solutions. But AI might change that.</p><p>Let’s go back to Jevons’ Paradox and apply it to software. AI tools are lowering the cost (in time and resources) of building software. Where it used to take weeks to build an MVP, you can now have a functional prototype in hours. Where a junior developer might have spent days learning an API, now they can ask their copilot to generate examples and explanations on the fly.</p><p>What will be the consequences?</p><ul><li><strong>Implementation costs go down</strong>: Thanks to AI, designing, building, and deploying software is faster and more accessible. This lowers the barrier for companies, entrepreneurs, and individuals alike.</li><li><strong>Demand increases</strong>: Many ideas that used to be unviable due to lack of budget, time, or technical staff will now become possible. The number of digital products will grow even more.</li><li><strong>Software won’t just be for engineers</strong>: Power users, designers, analysts, and hybrid profiles will be able to create their own tools. With AI’s help, non-technical people will automate tasks, build dashboards, or even create full business workflows.</li></ul><h4>Conclusion: More Software, Not Less</h4><p>No matter how much the tools change, the core of software development will still be essential. Someone will need to design, maintain, adapt, and improve these systems. Software engineers are not going away. Their work will evolve — as it has many times before.</p><p>AI is not the end of development. It’s the next evolution.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=54d7e0f831bd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[El lenguaje de programación más usado]]></title>
            <link>https://medium.com/@mbogarin/el-lenguaje-de-programaci%C3%B3n-m%C3%A1s-usado-446b509e4822?source=rss-7922852b0fd5------2</link>
            <guid isPermaLink="false">https://medium.com/p/446b509e4822</guid>
            <category><![CDATA[lenguajes-de-programación]]></category>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[tecnología]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[inteligencia-artificial]]></category>
            <dc:creator><![CDATA[Mario Bogarin]]></dc:creator>
            <pubDate>Mon, 30 Jun 2025 16:52:07 GMT</pubDate>
            <atom:updated>2025-06-30T16:52:07.305Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Ingleés como el nuevo lenguaje de programación" src="https://cdn-images-1.medium.com/max/1024/1*Mj1A6UTNAb1_GVWwIWHifg.jpeg" /><figcaption>Imagen de <a href="https://pixabay.com/es/users/generativestockai-18730327/">GenerativeStockAI</a> en <a href="https://pixabay.com/es/illustrations/ai-generado-c%C3%B3digo-binario-datos-8273245/">Pixabay</a></figcaption></figure><p>A estas alturas, no debería sorprendernos si afirmamos que el lenguaje de programación más usado en el futuro será (si no lo es ya) el inglés. O incluso tu propia lengua materna, como el castellano. No se trata de una metáfora, sino de una tendencia real que está transformando la forma en la que escribimos, diseñamos y mantenemos software.</p><p>Todo esto comenzó a tomar forma con un tuit viral de <strong>Andrej Karpathy</strong>, ex director de IA en Tesla y cofundador de OpenAI, donde afirmaba que <strong>el inglés es el lenguaje de programación más popular</strong> del momento. El tuit es de enero de 2023 y sigue siendo válido.</p><iframe src="https://cdn.embedly.com/widgets/media.html?type=text%2Fhtml&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;schema=twitter&amp;url=https%3A//x.com/karpathy/status/1617979122625712128&amp;image=" width="500" height="281" frameborder="0" scrolling="no"><a href="https://medium.com/media/0801f8c9b7b8002b309825e8ea50b9b2/href">https://medium.com/media/0801f8c9b7b8002b309825e8ea50b9b2/href</a></iframe><p>Su idea es clara: las habilidades de los modelos de lenguaje grande (LLMs) para comprender el lenguaje humano, combinadas con herramientas de desarrollo modernas, permiten que una parte importante del trabajo de ingeniería de software pueda realizarse usando lenguaje natural.</p><h4>Lenguaje natural como interfaz de programación</h4><p>Gracias a la madurez de modelos como <strong>GPT-4o</strong> y el auge de entornos como <strong>Cursor AI</strong>, <strong>Replit</strong>, <strong>Lovable</strong> o <strong>v0.dev</strong>, ahora es posible desarrollar, mantener y escalar aplicaciones usando inglés (u otros idiomas) como lenguaje principal de interacción. De hecho, en tareas como escribir código, estructurar proyectos, generar documentación, hacer debugging o refactorizar, muchas veces la instrucción en lenguaje natural reemplaza a líneas completas de código.</p><p>No todos los idiomas son iguales ante una IA, claro. Los modelos están entrenados con grandes cantidades de texto extraído de la web, libros, foros técnicos y documentación oficial, pero hay un sesgo natural hacia ciertos idiomas.</p><p>Así se estima la distribución aproximada de idiomas en el entrenamiento de un modelo como GPT-4o:</p><ul><li><strong>Inglés:</strong> 70–80% de los datos de entrenamiento</li><li><strong>Idiomas de alta disponibilidad </strong>(chino, español, francés, alemán…): 15–20%</li><li><strong>Otros idiomas: </strong>menos del 5%</li></ul><p>Estos datos no son públicos ni exactos, pero nos dan una visión aproximada de por qué el inglés sigue siendo el idioma dominante en interacciones con IA. Aun así, el castellano tiene cada vez mejor soporte, especialmente en tareas comunes de programación.</p><h4><strong>Código en inglés… literalmente</strong></h4><p>En una charla reciente para la <strong>AI Startup School</strong> de San Francisco, Karpathy amplió esta idea con un concepto clave: <strong>el código dejará de ser exclusivamente Python, JavaScript o C++</strong>. En su lugar, veremos más instrucciones en inglés directamente dentro de nuestros repositorios, archivos de configuración y flujos de trabajo.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FLCEmiRjPEtQ%3Fstart%3D684%26feature%3Doembed%26start%3D684&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DLCEmiRjPEtQ&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FLCEmiRjPEtQ%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/6e0e88fec69aeea762667b0f7f550c25/href">https://medium.com/media/6e0e88fec69aeea762667b0f7f550c25/href</a></iframe><p>El <strong>“Inglés como código gestionado” </strong>ya lo estamos viendo en prompts persistentes, pipelines definidos con texto natural, comentarios estructurados y hasta interfaces conversacionales que se integran con CI/CD o bases de datos. La barrera entre “escribir código” y “explicar lo que quieres hacer” está desapareciendo rápidamente.</p><h4>Vibe Coding y agentes inteligentes</h4><p>Este nuevo paradigma también está dando lugar a nuevos estilos de trabajo. Uno de ellos es el llamado <strong>Vibe Coding</strong>, en el que el desarrollador escribe ideas o intenciones (en lenguaje natural) y la IA se encarga de traducirlas en estructuras funcionales de código. Este estilo es colaborativo, conversacional y fluido. Muy diferente al paradigma tradicional de diseño, compilación y depuración manual.</p><p>Además, el auge de los <strong>agentes autónomos</strong> con capacidades de razonamiento y acción refuerza esta tendencia: <a href="https://medium.com/@mbogarin/qu%C3%A9-es-un-agente-de-ia-y-por-qu%C3%A9-va-a-cambiar-tu-forma-de-programar-db99ee67e490">ya no solo escribimos código para ser ejecutado, sino instrucciones para que otro sistema actúe por nosotros.</a></p><h4>Herramientas clave para programar con lenguaje natural</h4><p>A continuación, algunas de las herramientas más destacadas que permiten esta transición hacia un desarrollo más conversacional:</p><ul><li><strong>Replit</strong>: Un entorno de desarrollo colaborativo basado en navegador, con integración de IA y soporte para múltiples lenguajes. Permite generar, editar y ejecutar código directamente desde prompts.</li><li><strong>Lovable</strong>: Generador de productos digitales que convierte texto natural en MVPs funcionales (apps, webs, etc.). Ideal para diseñadores o makers sin conocimientos técnicos profundos.</li><li><strong>V0.dev</strong>: Plataforma que genera interfaces React + Tailwind a partir de descripciones escritas. Muy útil para prototipar UI rápidamente desde texto.</li><li><strong>Cursor</strong>: Editor de código derivado de VS Code, potenciado con IA. Su enfoque está en el desarrollo contextual, donde puedes pedir cambios y mejoras con frases en lenguaje natural.</li><li><strong>GitHub Copilot</strong>: Asistente de codificación que autocompleta código y sugiere funciones a partir de lo que escribes o describes. Integra profundamente con VS Code, Neovim, JetBrains y otros IDEs.</li></ul><h4>Bola extra: Automatización conversacional</h4><ul><li><strong>n8n</strong>: Herramienta de automatización open source que permite construir flujos lógicos (tipo Zapier). Algunos de sus módulos ahora permiten interacción en lenguaje natural con IA integrada, lo que facilita definir procesos sin necesidad de scripting.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=446b509e4822" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What Is an AI Agent and Why It Will Change How You Build Software]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@mbogarin/what-is-an-ai-agent-and-why-it-will-change-how-you-build-software-150ee5291d61?source=rss-7922852b0fd5------2"><img src="https://cdn-images-1.medium.com/max/1280/1*dYoLqw_MWoFVkZaroufEEA.jpeg" width="1280"></a></p><p class="medium-feed-snippet">Welcome to the future. Again. Since the 1980s, every few years someone announces the imminent extinction of the software developer. You&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@mbogarin/what-is-an-ai-agent-and-why-it-will-change-how-you-build-software-150ee5291d61?source=rss-7922852b0fd5------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@mbogarin/what-is-an-ai-agent-and-why-it-will-change-how-you-build-software-150ee5291d61?source=rss-7922852b0fd5------2</link>
            <guid isPermaLink="false">https://medium.com/p/150ee5291d61</guid>
            <category><![CDATA[agents]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Mario Bogarin]]></dc:creator>
            <pubDate>Thu, 26 Jun 2025 22:12:32 GMT</pubDate>
            <atom:updated>2025-06-26T22:12:32.721Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Qué es un agente de IA y por qué va a cambiar tu forma de programar]]></title>
            <link>https://medium.com/@mbogarin/qu%C3%A9-es-un-agente-de-ia-y-por-qu%C3%A9-va-a-cambiar-tu-forma-de-programar-db99ee67e490?source=rss-7922852b0fd5------2</link>
            <guid isPermaLink="false">https://medium.com/p/db99ee67e490</guid>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[agentes-de-ia]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[development]]></category>
            <category><![CDATA[inteligencia-artificial]]></category>
            <dc:creator><![CDATA[Mario Bogarin]]></dc:creator>
            <pubDate>Thu, 26 Jun 2025 21:51:35 GMT</pubDate>
            <atom:updated>2025-06-26T21:51:35.845Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dYoLqw_MWoFVkZaroufEEA.jpeg" /><figcaption>Imagen de <a href="https://pixabay.com/es/users/amrulqays-38834479/?utm_source=link-attribution&amp;utm_medium=referral&amp;utm_campaign=image&amp;utm_content=8563262">Amrulqays Maarof</a> en <a href="https://pixabay.com/es//?utm_source=link-attribution&amp;utm_medium=referral&amp;utm_campaign=image&amp;utm_content=8563262">Pixabay</a></figcaption></figure><p>Bienvenidos al futuro. Otra vez. Desde los años 80, cada cierto tiempo alguien anuncia la inminente desaparición del desarrollador de software. A saber: los lenguajes de alto nivel como C o Java iban a hacerlo todo más fácil, los frameworks lo automatizaban todo, el <em>no-code</em> venía a democratizar el desarrollo, la web mataba al escritorio, las apps móviles lo reinventaban todo. Y ahora… IA y agentes.</p><p>Cada ola tecnológica viene acompañada de titulares que prometen una revolución que nos dejará sin trabajo. Pero lo cierto es que, en lugar de extinguirnos, los desarrolladores hemos seguido aquí. Más ocupados que nunca, de hecho. Porque aunque las herramientas evolucionen, la demanda de software útil, adaptable y con sentido del negocio no ha parado de crecer.</p><p>¿Pero es esta vez diferente? ¿Lo son los agentes de IA? Vamos a ver.</p><p>En este artículo te explico qué es un agente de IA y por qué podrías tener que empezar a pensar en ellos como nuevos “usuarios” de tu software.</p><h4>Que es un agente de IA?</h4><p>La <a href="https://en.wikipedia.org/wiki/Agentic_AI">definición de wikipedia</a> dice algo como:</p><blockquote>La IA agente es una clase de inteligencia artificial que se centra en sistemas autónomos que pueden tomar decisiones y realizar tareas sin intervención humana.</blockquote><p>Mi version de esa definición es:</p><blockquote><strong>Un Agente de IA es una entidad basada en Inteligencia Artificial que, dado un objetivo, es capaz de diseñar y ejecutar un plan para cumplirlo</strong></blockquote><p>No estamos hablando de un simple chatbot que responde a preguntas, sino de algo más proactivo. Un agente es capaz de decidir <em>qué hacer</em> y <em>cómo hacerlo</em> con bastante autonomía.</p><p>Para entenderlo mejor, pensemos en sus piezas básicas:</p><ul><li><strong>Un objetivo</strong>: lo que el agente debe lograr. Por ejemplo, “monitoriza esta API y crea un informe diario”.</li><li><strong>Instrucciones o contexto</strong>: le dicen al agente cómo debería operar. ¿Debe priorizar la velocidad? ¿Evitar ciertos sistemas?</li><li><strong>Un LLM (Large Language Model)</strong>: el autentico cerebro del agente. como los que ofrecen OpenAI (o3, 4o, etc.), Google (Gemini), Anthropic (Claude) o Meta (Llama). Cuanto mas potente sea la LLM mas capacidad de entender el objetivo y actuar con sentio.</li><li><strong>Herramientas (tools)</strong>: piezas de software que el agente puede usar para hacer cosas. Desde llamar a una API hasta lanzar código o consultar bases de datos.</li><li><strong>Memoria</strong>: opcional pero poderosa. Un sistema de base de datos para guardar las decisiones pasadas y aprender para futuras ejecuciones.</li></ul><p>Y aquí es donde las cosas se ponen interesantes para nosotros, los desarrolladores.</p><h4>Un mundo con nuevos usuarios: los agentes</h4><p>Durante décadas hemos construido software pensando en humanos: pantallas, formularios, botones, clics, inputs predecibles, flujos de navegación… o, como mucho, en integraciones entre sistemas bien definidos con sus APIs documentadas, webhooks, SDKs, etc.</p><p>Pero ahora aparece un nuevo tipo de “usuario” en el vecindario: <strong>el agente de IA</strong>.</p><p>Y este usuario no usa ratón ni teclado. Llama funciones, ejecuta comandos, interpreta descripciones. Actúa con libertad. A veces hace cosas que ni tú habías previsto.</p><p>Esto cambia las reglas del juego.</p><p><strong>¿Qué implica esto para ti como dev?</strong></p><p>El auge de los agentes de IA va a generar, como mínimo, tres áreas de trabajo clave::</p><ul><li><strong>Infraestructura de agentes</strong>: Tendremos que pensar en cómo escalar estos sistemas, cómo observarlos, auditar su comportamiento y asegurar que no hagan cosas raras. Es decir, observabilidad, logging, seguridad, límites de ejecución, etc… como con cualquier otro sistema complejo.</li><li><strong>Ingeniería de prompts</strong>: Las instrucciones que le des al agente (los prompts, las descripciones de tareas, etc.) van a requerir revisión, pruebas, validación, y gobernanza.</li><li><strong>Creacion de herramientas para agentes:</strong> e trata de construir la nueva capa de abstracción que los agentes puedan usar. Librerías, endpoints, funciones bien definidas y robustas, pensadas no para humanos, sino para una IA que las va a invocar y combinar de maneras creativas. Frameworks como MCP o A2A ayudarán con la estandarización.</li></ul><h4>¿Y ahora qué?</h4><p>Los agentes de IA no eliminan a los desarrolladores. Pero sí cambian el terreno de juego. Nos dan nuevos retos, nuevos tipos de usuarios, nuevas capas de abstracción por construir. Y, si lo piensas bien, eso siempre ha sido parte del trabajo.</p><p>Bienvenidos, Agentes.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=db99ee67e490" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Who’s Driving? AI, Translation, and Software Development]]></title>
            <link>https://medium.com/@mbogarin/whos-driving-ai-translation-and-software-development-abdd84c37c61?source=rss-7922852b0fd5------2</link>
            <guid isPermaLink="false">https://medium.com/p/abdd84c37c61</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[llm-applications]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Mario Bogarin]]></dc:creator>
            <pubDate>Wed, 25 Jun 2025 20:23:18 GMT</pubDate>
            <atom:updated>2025-06-25T20:23:18.110Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XbYPN-xv3JSmXxy7H50Okw.jpeg" /><figcaption>Picture from <a href="https://pixabay.com/es/users/roketpik-5186989/?utm_source=link-attribution&amp;utm_medium=referral&amp;utm_campaign=image&amp;utm_content=5113374">Artur Shamsutdinov</a> at <a href="https://pixabay.com/es//?utm_source=link-attribution&amp;utm_medium=referral&amp;utm_campaign=image&amp;utm_content=5113374">Pixabay</a></figcaption></figure><p>In the localization industry, there’s been a clear trend for years now, and it’s only accelerated with the rise of generative AI: <strong>post-editing</strong>. If you’ve worked in translation, you’ve probably come across the term <strong>PEMT</strong> (Post-Editing Machine Translation). Basically, the translator no longer starts from scratch — they receive a text that has already been machine-translated, and their job is to fix, polish, or sometimes completely rewrite parts of it.</p><p>Sounds simple enough, right? But there’s a deeper shift happening here. Translation used to be a creative process you could bring your voice, your style. Now, with post-editing, the task often becomes more mechanical. Instead of building from the ground up, sometimes you’re reviewing and correcting what someone (or something) else produced. That changes the relationship you have with your work.</p><p>So, what does all this have to do with software development? A lot more than it might seem at first. In fact, it’s a great example of how <strong>AI is reshaping the workflows</strong> of skilled professionals in many technical fields.</p><p>Because let’s face it: what’s happening in translation is starting to happen in code. With tools like <strong>Cursor AI</strong>, <strong>GitHub Copilot</strong>, <strong>Replit</strong>, <strong>Devin</strong>, and many others, the role of the developer is shifting. We’re not being replaced. We are evolving. Instead of writing every line ourselves, we’re increasingly becoming reviewers, editors, and ultimately, the ones responsible for the final result of what an AI generates.</p><h3>Who’s in the Driver’s Seat?</h3><p>If you’ve ever played around with Cursor AI or a good LLM-powered tool, chances are you’ve had one of those “<strong>wow</strong>” moments. You type in a prompt like:</p><blockquote>“Create an API in Go with layers: route, handler, service, and repository. Start with a simple endpoint to show a list of users.”</blockquote><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F3U5haQk5uO4%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D3U5haQk5uO4&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F3U5haQk5uO4%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="640" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/d9a14ea6cef0a99b144e36553ebb32e4/href">https://medium.com/media/d9a14ea6cef0a99b144e36553ebb32e4/href</a></iframe><p>And boom: in a few seconds, you’ve got a full backend scaffold. Routes, services, handlers, even a few comments thrown in. It’s pretty impressive.</p><p>But then you look closer… and reality sets in. Is it well-structured? Is it idiomatic? Are there subtle bugs or edge cases? Does it match the actual needs of your product? Spoiler alert: not always.</p><p>That’s where you come in. Because even though the AI can write surprisingly decent code, <strong>you’re still the one who puts it into production</strong>. You decide whether it stays, gets tweaked, or goes straight into the trash. The job isn’t just about “accepting” the output. It’s about <strong>post-editing</strong> it — just like in translation.</p><p>And that takes judgment, experience, system knowledge, an understanding of the business context, and of course, the ability to spot what’s going on behind the scenes. What looks like magic is often just a starting point.</p><h3>From Keyboard to Prompt Editor</h3><p>This new reality doesn’t mean developers are going away. If anything, we’re more necessary than ever. But the nature of the job is shifting. It’s less about typing every character , and more about <strong>guiding the AI</strong>, <strong>validating its output</strong>, and knowing how to build real software from AI suggestions.</p><p>Just like a good translator doesn’t just fix typos, a good developer doesn’t just compile what Copilot gives them. You need to understand, adapt, improve, and sometimes completely rewrite.</p><p>The big question is: <strong>how do we adapt?</strong> Because if you’re not in the driver’s seat, someone — or something — else will be. And that “someone” might not understand your users, the business, the architecture, or the security risks. Bottom line: the <strong>copilot is not the pilot</strong>. Not yet, anyway.</p><p>So yes, AI is changing the way we work. But as long as we keep the judgment, the perspective, and the control — <strong>we’re still the ones driving</strong>.</p><p><em>Note: This is a translation from the </em><a href="https://medium.com/@mbogarin/qui%C3%A9n-conduce-ia-traducci%C3%B3n-y-desarrollo-de-software-1a4704766402"><em>original spanish text</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=abdd84c37c61" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[¿Quién conduce? IA, traducción y desarrollo de software]]></title>
            <link>https://medium.com/@mbogarin/qui%C3%A9n-conduce-ia-traducci%C3%B3n-y-desarrollo-de-software-1a4704766402?source=rss-7922852b0fd5------2</link>
            <guid isPermaLink="false">https://medium.com/p/1a4704766402</guid>
            <category><![CDATA[llm-applications]]></category>
            <category><![CDATA[inteligencia-artificial]]></category>
            <category><![CDATA[programación]]></category>
            <category><![CDATA[desarrollo-de-software]]></category>
            <dc:creator><![CDATA[Mario Bogarin]]></dc:creator>
            <pubDate>Wed, 25 Jun 2025 17:57:47 GMT</pubDate>
            <atom:updated>2025-06-25T17:57:47.448Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XbYPN-xv3JSmXxy7H50Okw.jpeg" /><figcaption>Imagen de <a href="https://pixabay.com/es/users/roketpik-5186989/?utm_source=link-attribution&amp;utm_medium=referral&amp;utm_campaign=image&amp;utm_content=5113374">Artur Shamsutdinov</a> en <a href="https://pixabay.com/es//?utm_source=link-attribution&amp;utm_medium=referral&amp;utm_campaign=image&amp;utm_content=5113374">Pixabay</a></figcaption></figure><p>En la industria de la localización hay una tendencia bastante clara desde hace años, y se ha acelerado desde la irrupción de la IA generativa: la post-edición. Si has trabajado en traducción, seguro que te suena el término <strong>PEMT</strong> (Post-Editing Machine Translation). Básicamente, consiste en que el traductor ya no empieza de cero, sino que recibe un texto previamente traducido por una máquina, y su tarea es corregirlo, pulirlo o directamente reescribir partes.</p><p>Visto así, parece sencillo. Pero hay algo más profundo detrás de este cambio. Tradicionalmente, traducir era una tarea creativa, donde podías dejar tu estilo, tu voz, tu forma de entender el texto. Con la post-edición, muchas veces la cosa se vuelve más mecánica. En lugar de crear desde cero, pasas a revisar, arreglar y validar el trabajo de otro — en este caso, de una máquina. Y eso, queramos o no, cambia la relación que tienes con tu oficio.</p><p>Ahora bien, ¿qué tiene que ver todo esto con el desarrollo de software? Pues más de lo que parece. De hecho, es un ejemplo bastante claro (y útil) de cómo la Inteligencia Artificial está cambiando la forma en que trabajamos los profesionales técnicos y cualificados.</p><p>Porque si te fijas, lo que está pasando en la localización también está empezando a pasar en el código. Con la llegada de herramientas como <strong>Cursor AI</strong>, <strong>GitHub Copilot</strong>, <strong>Replit</strong>, <strong>Devin</strong>, entre otras, el rol del desarrollador empieza a transformarse. No desaparece — ojo, que esto no va de reemplazos mágicos — pero sí evoluciona. Ahora, más que escribir cada línea desde cero, muchas veces te conviertes en revisor, editor y responsable final de lo que una IA propone.</p><h4>¿Quién está al volante?</h4><p>Si alguna vez has usado Cursor AI o una LLM bien afinada, seguro que has tenido ese momento “<strong>wow</strong>”. Esos segundos en los que escribes un prompt del estilo:</p><blockquote>Create an API in go with layers: route, handler, service and repository. Start with a simple endpoint to show a list of users</blockquote><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F3U5haQk5uO4%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D3U5haQk5uO4&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F3U5haQk5uO4%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="640" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/d9a14ea6cef0a99b144e36553ebb32e4/href">https://medium.com/media/d9a14ea6cef0a99b144e36553ebb32e4/href</a></iframe><p>Y zas: en cuestión de segundos, tienes una estructura de backend básica. Rutas, servicios, handlers, todo bien ordenado y con comentarios. Te dan ganas de aplaudirle al modelo.</p><p>Pero luego lo miras con calma… y empieza la realidad. ¿Está bien estructurado? ¿Está usando buenas prácticas? ¿Tiene errores sutiles o edge cases mal resueltos? ¿Cumple con los requisitos reales del producto? Spoiler: casi siempre, no del todo.</p><p>Ahí es donde entras tú. Porque aunque la IA sea capaz de escribir código sorprendentemente decente, tú eres el que lo lleva a producción. Tú decides si eso se queda, se ajusta o se tira y se empieza de nuevo. El trabajo no es simplemente “aceptar” el output, sino <strong>posteditarlo</strong>. Justo como en el mundo de la traducción.</p><p>Y eso implica criterio, experiencia, conocimiento del sistema en su conjunto, entender el contexto del negocio y, por supuesto, saber qué hay detrás del código que estás revisando. Lo que parece magia es, muchas veces, solo el punto de partida.</p><h4>Del teclado al editor de prompts</h4><p>Este nuevo escenario no significa que los desarrolladores vayamos a desaparecer. Más bien lo contrario: vamos a ser más necesarios que nunca. Pero el foco está cambiando. Ya no se trata solo de picar líneas de código a mano, sino de entender mejor cómo guiar a la IA, cómo validar lo que propone y cómo construir software de verdad a partir de esas sugerencias.</p><p>Igual que un buen traductor no se limita a corregir errores gramaticales, un buen desarrollador no se limita a compilar lo que le da Copilot. Hay que entender, adaptar, mejorar y, cuando hace falta, reescribir.</p><p>La gran pregunta es: ¿cómo nos adaptamos? Porque si tú no estás al volante, alguien más lo estará. Y ese alguien podría no entender el negocio, no conocer a los usuarios, no tener en cuenta la escalabilidad o la seguridad. En resumen: el copiloto no es piloto. Aún no.</p><p>Así que sí, la IA cambia la forma de trabajar. Pero mientras nosotros mantengamos el criterio y el control, seguimos siendo los que llevamos las manos en el volante.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1a4704766402" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>