I tend to agree completely with this and this is always a warning I issue to all beginners. If your plan is to master both the technology and the process of delivering reliable software in reasonable time, the only way to do that is by getting your hands dirty. LLMs will boost your output significantly when you are just starting out, however that is more or less the final ceiling of your performance, in 1-2 years you will be left behind in the dirt by the folk that decide to understand the problem fully before solving it.
I’ve worked some time ago (pretty much right when chatGPT was released) with a few developers that had about 2 years less experience than me and I was horrified at the end of the project. They were relying solely on prompts to do everything and while their performance was in some places better, the lack of direction and understatement of how the problem at hand can be approached made the project take longer than it would have if they were just not using the LLM. The main reason behind that is that they didn’t try to solve the problem but wanted the LLM to do that and the effect is that they introduced a lot of unnecessary external noise which made the process much longer and unnecessary complex.
Just like any other things, this is just a tool. Understanding when and where to use it is crucial, and most developers are simply horrible at assessing this. This dates from long ago and it’s no different if we speak about tools that came before it.
While I completely agree that we need to put food on the table, I would also say that the only way to win this is by not playing the game. I think we overestimate how important is the skill of writing fast code is, code should be considered a liability not an asset. A well solved problem will beat by at least a factor of 10x any kind of artificial boost, be it code generation or something else. I can say that this is certainly the case when it comes to elixir, there are countless small companies that can have 3-10 engineers over the lifetime of the company and are beating the competition to a pulp, and trust me those engineers spend less than 20% of their time writing code.
There is and always will be, but not for factory workers that are interested in typing code all day long. There is an extreme shortage of folk that can solve business problems and deliver a working solution, which is the hardest skill to learn and master (and will never be replaced by a LLM, as this is the tipping point where actual intelligence is required). I’ve personally had the luck to be thrown into the fire in starting and developing some amazing and complex products and I can say that I’ve failed miserably time after time to come with good solutions, however that taught me some of the most valuable lessons that made me see the profession of a developer as something different.
I think that the start of the gaming development industry symbolizes how being capable of solving problems is more important than anything else. I would highly recommend to watch Half-Life: 25th Anniversary Documentary, there is a point in that video where everyone points that most of the folk that worked on making the game were coming from other trades, they learned and created new tools and managed to deliver an amazing game.
