Digital obesity in the age of AI

Have you ever wondered why your phone gets scorching hot after using an app that, five years ago, ran on a device with a tenth of the power?

Well, perhaps what is happening in your pocket is also happening at the heart of your company: technology investment is evaporating into layers of fat. If you think AI is the miracle diet that will cure this 'digital obesity', brace yourself, because we may simply be adding more bulk to the problem. Or maybe not ...

A brief history of programming ... and of the role abstraction has played in it

For those of you who are not part of the tech industry, the concept of a "programming language" may sound like some strange arcane art practiced by peculiar people who struggle to communicate with normal humans. That, at least, is the summary most of the normal people I know have formed in their heads. But it is an inaccurate notion, and without grasping it, it is hard to understand what is happening in the world today.

To cut a very long story short, every computing device is driven by a processor that executes instructions written in something called "machine code".

• In the 1940s, programmers wrote this code directly: complex, and reserved for a handful of initiates.
• Assembly language appeared to pass direct instructions to the processor (one more level of abstraction).
• In the '70s and '80s we moved on to "imperative languages" (C or Pascal), where the focus was on logic and resource management (memory).
• Then came "object-oriented" languages, allowing programmers to concentrate on modeling the business.
• In the 2010s, taking abstraction one step further yet again, the era of frameworks and the cloud began, with the focus on rapid deployment and the assembly of services.

Today we are in the age of "vibecoding", AI-assisted programming in natural language, or whatever you want to call it. The latest invention of the wise Jews of Amsterdam, as García Márquez would have put it. Thanks to it, anyone can "program" simply by describing to an AI what the application they need looks like.

Are "vibecoding", "low code" or "no code" really a disruption?

A full-blown disruption, no doubt about it. The fact that to build an application I no longer have to call in a "techie" and try to establish a common language in which he understands my business idea and I understand his programming concepts is wonderful, isn't it?

Well, yes and no. This new generational leap in the history of programming has tremendous disruptive potential, but like everything in life, we need to understand it if we want to make the most of it. Above all, one idea must be crystal clear: in the end, beneath an enormous stack of abstractions that progressively distance us from the hardware, there will always be a processor executing instructions in machine code.

In other words: being able to do without the techie does not come for free. And that is why your phone overheats and your company's cloud computing budget grows voraciously every year.

Who benefits from this new scenario?

A world where anyone can program is, in itself, an interesting scenario. Think of companies that will no longer need a techie or a consultancy to develop custom applications for them, of individuals who will be able to build their own app to support a hobby, of business school students who will be able to generate on their own the applications in which their business ideas used to take shape, or of gamers who will be able to design the next generation of online games without needing programming studios.

And all of this with tools that are free or nearly free? How is that possible?

Seneca said that he who profits from the crime is the one who committed it. If we pull on that thread, it doesn't take much imagination to work out who stands to benefit from mountains of amateurs generating what they don't realize is doped-up code. Because AI-generated code tends to use extremely heavy libraries to solve simple problems, simply because that is the statistically most likely option. And they build you an 18-wheeler to go and buy the Sunday bread. But ... what's the problem? Generating 10 lines of code will cost me the same as generating a million!

Well, setting aside ethical or environmental considerations, the problem with this obese software is that it will need enormous computing capacity to solve simple problems. And that is why Big Tech will keep handing us tools so that anyone can program. They'll send us the bill when the time comes to run that code and we realize it wasn't all as rosy as it seemed.

Because the cost of creation (which will drop substantially) is one thing, and the cost of execution (which will skyrocket the more frequently a piece of code is used and the higher the level of abstraction from which it was generated) is quite another.

I know it sounds harsh, but in this little world the business model is starting to look dangerously like that of the drug trade: the industry gives you your first doses of ease and speed for free so that, once you're hooked on an ecosystem of obese software you no longer even know how to maintain, you have no choice but to pay whatever price they set for the energy needed to run it. We have traded creative freedom for dependence on infrastructures that are in the hands of a few.

Should technology consultancies be worried?

I'm not hiding. I work for a consultancy, Sngular, and from here we view this scenario with a mixture of fascination and respect. In our internal forums, this is not a theoretical debate; it is our daily bread. In fact, these lines were born precisely from a conversation with a colleague about how these tools are altering levels of abstraction and, with them, the future of our profession.

I won't hide from you that a movement as disruptive as the one these new tools are bringing is quite a challenge. Because not every client is aware that they are living through a shift in cost: from "programmers' salaries" to the "computing bill". And of course, because there are also consulting firms whose focus has been to increase, as much as humanly possible, the "man-hours" they bill their clients. But I also believe that development companies that focus on selling architectures to solve complex problems are going to have more work than ever. Especially if they are also capable of trimming all the fat generated by non-technical people modeling systems with AI tools.

The future of our profession seems to be splitting in two: those who generate volume and those who know how to reduce it. Are we ready to be the 'dietitians' of AI-bloated software, or will we settle for continuing to add layers of fat to the system?

Remember that, as the proverb goes, easy roads never lead far.