The era of accessible custom software is here

And the question isn’t whether it will affect you, but what you’re going to do about it.

---

Five years ago, if you had an idea for an application that solved a specific problem in your business — say, a tool for your sales team to log quotes with the exact logic of your discount structure, or a dashboard that showed you in real time how much material you had left in the warehouse cross-referenced with the week’s orders — you had basically two options. The first: hire a software development company, wait weeks or months, pay tens of thousands of dollars, and cross your fingers that the result resembled what you’d imagined. The second: give up. Adapt to generic software that sort of works. Learn to live with spreadsheets that break when someone accidentally deletes a formula.

Those two options are no longer the only ones. And what replaced them isn’t an incremental improvement — it’s a paradigm shift.

What changed (and why it matters if you’re not a programmer)

In February 2025, Andrej Karpathy — co-founder of OpenAI and former head of artificial intelligence at Tesla — published a tweet that was seen more than four million times. It said something like: “There’s a new kind of programming that I call vibe coding. You fully give in to the vibes, embrace the exponentials, and forget that the code even exists.” The phrase was deliberately informal, almost irreverent. But it captured something real: that for the first time in the history of computing, creating functional software no longer requires knowing how to code. It requires knowing how to describe what you need.

Collins Dictionary named vibe coding “Word of the Year 2026.” Not because of a trend, but because the term encapsulates a transformation already underway: today, 92% of developers in the United States use artificial intelligence tools every day. 41% of all code written in the world is generated by AI. GitHub Copilot — one of the most popular assistants — has more than 20 million users and generates over two billion dollars a year.

But the figure that should matter most to you, as a non-programmer, is a different one: at Y Combinator — the world’s most influential startup accelerator — 25% of the companies in its Winter 2025 cohort had codebases generated more than 95% by artificial intelligence. One in four ventures from the most important unicorn factory on the planet was built almost entirely by machines that receive instructions in English. Or in Spanish. Or in any language.

”Software for one”: the revolution nobody told you about

Kevin Roose, a journalist at the New York Times, coined a phrase that should be on the wall of every entrepreneur’s office in Latin America: “software for one.” It refers to applications that aren’t designed for millions of users. They don’t aim to scale. They don’t need investors. They are tools tailor-made for exactly one person — or one team, or one particular business — to solve a specific problem that no commercial software solves well.

Roose, who is not a programmer, built several of these apps using artificial intelligence tools. One of them analyzed the contents of his refrigerator and suggested what to pack for lunch. It sounds trivial. But think about the logic behind it: an Australian doctor, also with no programming experience, used the platform Replit to build in two or three days an app that verifies whether a patient’s treatment plan complies with clinical guidelines. Another that lets his patients complete a checklist before visiting him. Yet another that sends personalized health tips to his patients’ phones.

Two or three days. Without knowing how to code. Apps that would have previously required a full development team.

Now translate this to your context. The construction company that needs an app for its residents to report defects from their phones with geolocated photos. The accounting firm that wants a smart form that preloads the client’s tax information and detects inconsistencies before filing the return. The restaurant chain that wants its own system for managing shifts cross-referenced with historical demand by day of the week. Each of these is a case of “software for one” — solutions that the market will never build because they’re too specific for a generic product. And yet, they are exactly the tools that make the difference between a business that operates with friction and one that flows.

But here comes the uncomfortable part

If everything up to this point sounds like a Silicon Valley commercial, it’s because the other half of the story is missing. And that half is the one that can save you — or sink you — depending on how well you understand it.

Artificial intelligence has already democratized the ability to write code. That is a fact. Anyone with a clear idea can sit down in front of Claude, ChatGPT, Cursor, or Replit and get a functional application in hours. But — and this is critical — writing code has never been the same as building software.

A 2025 report from Veracode found that 45% of code generated by artificial intelligence fails security tests. Fast Company projects $1.5 billion in accumulated technical debt by 2027, caused in large part by AI-generated code without oversight. Kevin Roose, the same journalist who celebrates “software for one,” documented a case where AI-generated code fabricated fake reviews for an e-commerce site. The AI didn’t do it out of malice — it did it because nobody told it not to.

The German technology blog Ströer summed it up with an elegance worth repeating: “AI tools have democratized programming, but they have not abolished responsibility for sustainable code. On the contrary: they have distributed it across more shoulders.”

The truth is that software development was never just about writing code. That was merely the visible manifestation. Beneath the code there was always something harder, more human, more valuable: understanding a business’s processes. Grasping its culture. Making decisions based on information that can’t always be expressed in words — intuitions about what works and what doesn’t, accumulated experience about how users react, tacit knowledge about exceptions and edge cases that no requirements document fully captures.

What AI cannot do (yet)

Think of it this way: AI can write you an application that logs your sales team’s quotes. But it doesn’t know that your best salesperson, Laura, always rounds discounts to the nearest even number because they “look more professional” in the proposal. It doesn’t know that your approval process changes in December because everything accelerates before year-end. It doesn’t know that the client who asks for “something simple” actually wants you to solve five things they didn’t mention because they assume they’re obvious.

AI doesn’t know that your inventory system needs to connect with the ERP you bought six years ago that runs a version nobody supports anymore. It doesn’t know that your warehouse team is intimidated by computers and that if the interface has more than three buttons they’ll go back to the notebook they’ve always used. It doesn’t know that your regulator changed a form format last week and that if the report doesn’t come out exactly right, you get fined.

Klaus Haeuptle, a software engineer and author of the newsletter Engineering Ecosystem, posed the question directly in 2025: does AI really democratize software development? His answer: “It depends.” For prototypes, quick internal tools, automating repetitive tasks — yes, absolutely. For software that needs to be maintained long-term, that has to be secure, that must integrate with existing systems, that needs to scale as the business grows — no. There, you still need people who deeply understand how software works, how your business works, and how to make both work together.

The metaphor that best captures this is the 80/20 rule: vibe coding can produce the first 80% of a project in hours. The last 20% — the part that makes it secure, maintainable, integrated, scalable, and genuinely useful in the real world — still requires experience, judgment, and human guidance.

Why is this especially urgent in Latin America?

There is a data point from RAND Corporation that should be on the front page of every business newspaper in the region. In an analysis published in October 2025, RAND noted that language models have eliminated English as the gatekeeper of software innovation. In their words: “With LLMs now fluent in virtually every language in the world, coding and application development can thrive in Chinese just as easily as in English.” What RAND wrote with China in mind applies with equal force to Latin America. Today, you can describe your software in Spanish — with all the specificity of your context, your industry, your market — and AI builds it.

The language barrier that for decades kept the Spanish-speaking world one step behind in the software revolution has just disappeared.

According to a study by NexoEstelar, 77% of small businesses worldwide already use some form of artificial intelligence tool. In Mexico, the Tecnológico de Monterrey developed TECgpt — its own generative AI platform — with an explicit low-code approach, designed so that the university and business community can adopt it without needing to be experts. Latin American companies like Nubank, Mercado Libre, Rappi, and Bancolombia are already using AI in ways that would have been science fiction five years ago. And contrary to popular fear, the data show that 13.7% of companies that adopt AI increased their headcount, compared to only 6.9% that reduced it.

But the most revealing numbers don’t come from the giants. They come from the SME that automated its customer service with a chatbot built in an afternoon. From the architecture firm that stopped wasting hours searching for versions of blueprints because someone created an internal search engine with AI. From the auto parts distributor that connected its catalog with an assistant that answers technical questions from its customers via WhatsApp. Those stories don’t make the headlines yet, but they are happening right now, in cities across the region.

What the world is doing (and what you can learn from it)

Every region of the world is responding to this revolution from its own perspective. And there are valuable lessons in each one.

Japan approved its AI Promotion Act in May 2025, with the stated goal of becoming “the most AI-friendly country in the world.” But the interesting thing about Japan is not how much it invests — more than $66 billion committed by 2030 — but how it thinks about integration. Japan doesn’t aim to dominate foundational models. What it seeks is to demonstrate how technology can be meaningfully integrated into society — into its manufacturing, its robotics, its embedded systems. Its bet is on intelligent implementation, not the race for the biggest model.

China shook the world in January 2025 when DeepSeek launched a high-performance model at a fraction of the cost of its Western competitors. In 2025, China surpassed the United States in volume of AI patents. Its aggressive bet on open-source aims to transform it not only into a manufacturing powerhouse but into a software powerhouse — a domain where it had historically lagged behind.

Germany, true to its character, approaches the topic with enthusiasm tempered by caution. The DeKIOps project at the Fraunhofer Institute works specifically on making artificial intelligence tools accessible to industrial workers without technical AI knowledge. But the German voice that resonates most is that of a Ströer corporate blog that wrote what should be the mantra of this era: “Between ‘perfect or nothing’ and ‘as long as it works,’ there is a third path: consciously imperfect, but responsible.”

And Latin America? Our region doesn’t need to invent the models. It doesn’t need to compete with OpenAI or DeepSeek over who has the most powerful LLM. What it needs — and what it has a historic opportunity to do — is to adopt these tools intelligently, with local context, and at the speed the moment demands.

So, what is the real opportunity?

The opportunity is not that “now anyone can program” — although technically that’s true. The opportunity is that the cost and complexity of creating software tailored to your business have been reduced by an order of magnitude. What used to cost $50,000 and six months can now be achieved for a fraction of that cost and in weeks. What used to require a team of ten engineers can now be done by a team of three, because AI handles the mechanical writing of code and humans focus on what truly matters: understanding your business, designing the right solution, making sure it works in the real world, and iterating when surprises inevitably arise.

And that is what sets this era apart from the earlier promises of “no-code” and “low-code” that never fully delivered on what they offered. It’s not that technology has eliminated the need for expertise. It’s that expertise can now focus where it generates the most impact. Code became a commodity; what holds value is judgment, context, business knowledge, and the ability to ask the right questions.

Think of it from your company’s perspective: until yesterday, custom software was a luxury reserved for those who could afford the fees of a large consulting firm or were lucky enough to find a competent freelance developer. Today, the conversation changes. Today, you can sit down with a team that understands both technology and business, describe your problem in your own words, see a prototype in days, and have a working solution in weeks. Not because complexity has been eliminated, but because AI tools absorbed the mechanical part and freed people to focus on what’s strategic.

The time to decide is now

There is a window of opportunity that won’t last forever. Right now, most SMEs in Latin America still operate with generic tools, manual processes, and the same spreadsheet someone created in 2018 that nobody dares touch anymore. Meanwhile, the competitors who adopt these tools first — whether to automate their operations, to create superior customer experiences, or simply to stop wasting hours on work that a machine can do better — will build advantages that compound over time.

I’m not talking about replacing your team with AI. The data show that companies adopting artificial intelligence tend to grow their teams, not shrink them. I’m talking about giving your team the tools it needs to perform at the level the 2026 market demands.

The era of accessible custom software is already here. It’s not a promise of the future. It’s a fact of the present. What’s missing is for more companies — your company — to decide to walk through that door.

And the good news is that walking through that door has never been more viable. AI will write the code. But your business needs something AI still cannot offer: someone who understands why that code matters, who it works for, and how to scale it when success arrives. That combination — the speed of AI with the depth of human experience — is the real revolution. And it’s available today.

The question is whether you’ll be among those who seized it in time, or among those who kept reading about it.

---