This is not another post about the developer apocalypse
~ cat "This is not another post abo"~ cat post <<
Hundreds of professions have been completely obliterated since the human brain got smart enough to use a rock shard to defend itself. Today, a fan is so cheap that almost everyone can afford one. Hundreds of years ago, to stay cool, you had to be rich to hire someone to wave a giant fan while you did whatever you wanted.
The creation of LLMs first created a mesmerizing feeling, but also brought back prophets of newly doomed professions. They’re not wrong: many professions will die, as many have before, and some should already be extinct. IT professions are currently marked for obsolescence, but the people predicting their demise are often the same ones selling AI and LLM solutions.
However, this post isn’t about the downfall of any profession; it’s about a paradigm shift.
Modern fans are not only less expensive than old-school human fans but also far more efficient, powerful, tireless, and privacy-preserving. A human fan doesn’t make sense anymore.
But programmers? I’m sorry, you’re free to dislike them, but you can’t get rid of them: a well-prepared person with a machine is better than a machine alone, and there are many things they can do together.
The Human in the Loop: Augmentation, Not Replacement
Although writing software tests is good practice, many companies, especially smaller ones, don’t write them or write them poorly. Why? Writing good tests to cover all happy and sad paths is boring and expensive. Good tests require more lines of code than the tested code itself, and with limited budgets and time, it’s a real problem. On the other hand, using AI to detect potential bugs and implement automated tests is awesome.
Even though they can generate automated tests, LLMs will always hallucinate: they create tests that don’t make sense and sometimes claim the tests are faulty and should be adjusted to fit the function’s results. This is where developers shine: they can evaluate these issues and make the proper interventions. This human oversight is crucial not just for correcting errors, but for managing the scale of a project.
To put it another way, let’s suppose you, a regular person, decide to build a house tile by tile. It will take you months and cost more than you can afford. But when you see a house, your brain instantly recognizes it as a house. The same applies to a skyscraper, a neighborhood, or an entire city, and it applies to many things LLM AIs can do. While they can build fast, we can quickly check if it works.
Those used to LLMs for development know: you must request small portions of code, or you have to write lengthy prompts to specify all restrictions. The latter is a waste of time, while the former is faster than writing every line of code.
