Vibe Coding

In today's digital ecosystem (especially on Twitter, my natural habitat), the term "vibe coding" has gained popularity. This concept describes the act of "programming" by giving instructions to AI systems like ChatGPT or Claude. The question that immediately arises is: can we genuinely consider this programming?

Source of the cover picture.

Speed as a Competitive Advantage

We live in an era where speed defines success. Velocity determines whether you'll secure funding for your project or how quickly you can implement significant changes in your environment. Those who have integrated AI into their workflows can launch products ("just ship it!") with greater agility than those who haven't yet adopted these tools.

The Human Response to Innovation

Every technological change generates uncertainty and fear—a deeply human response to the unknown. This phenomenon provokes three main reactions:

1. The defenders of the status quo will question any disruptive change

2. The early adopters will incorporate novelty to "hop on the bandwagon" and "stay up-to-date" (often without real necessity, as novelty doesn't always represent the optimal solution)

3. The strategic adapters will explore new possibilities without disconnecting from the fundamentals that made them possible

This third group will likely navigate the waters of technological change more successfully.

Optimized Tools vs. Versatile Tools

I was recently discussing with colleagues a LinkedIn post comparing the performance of an LLM versus a calculator for multiplication tasks. This comparison illustrates a crucial point: each tool has its specific purpose.

Calculators are optimized for arithmetic calculations (there's a fascinating history about their development at the Arithmeum museum in Bonn, Germany). However, asking a calculator to write a paragraph about artificial intelligence would be pointless—it wasn't designed for that task.

Caption: Johann Christoph Schuster, 1820/22, The last artifact: circular calculating machine with steped drums, © Arithmeum.

Similarly, although an LLM can generate coherent texts with relative ease, its performance in precise calculations will be deficient compared to a specialized calculator.

The True Potential of AI

What's truly interesting emerges when we consider calculations so complex that even modern calculators and computers struggle to solve them. In these cases, it's sometimes necessary to resort to simulation to obtain a sufficiently good result—as happens with NP-type problems, exemplified by the traveling salesman problem, or non-linear dynamic systems used in meteorology, which lack definitive analytical solutions.

Caption: Solution of a travelling salesman problem: the black line shows the shortest possible loop that connects every red dot. Source: Wikipedia.

An LLM that can determine when I need a calculator and when I need a simulation—that contextual discernment capability—sparks my curiosity much more than evaluating its ability to perform basic arithmetical tasks.

But to fully leverage this potential, we must remember the fundamentals and understand how each component of the system works. "Vibe coding" may represent a new frontier, but only those who understand both the cutting edge and the foundations that support it will be able to harness its true potential.