AI and the future of programming

Starting from basic numerical models, the capacity of AI has grown tremendously; it has been able to process and create images, generate realistic texts, create deep fakes, and so much more. Despite this, though, AI in its current state can’t replace programmers. Here are a few reasons why:

• AI can’t understand complexities in more advanced code. Although it can comprehend code to a basic level and has some knowledge of higher-level intricacies, it can’t code a complex project well. As an anecdote, I tried to create a Scrabble program using ChatGPT, but it only took me part of the way there. Understanding how to link code together and finding mistakes in complex sections of code is much more difficult for LLMs.
• Although some AIs are good at math, LLMs like ChatGPT and Claude are subpar at math, and can even mess up basic addition and subtraction. Although math isn’t directly needed to code, understanding complex math is occasionally useful for complex programs, and AI cannot help in this regard.
• AI works like a computer - it gets things done in the minimum way possible. Trying to solely rely on AI for large-scale projects won’t work because it will create a bare minimum-level structure. Oftentimes, trying to adapt this code for more complex tasks is exceedingly difficult (speaking from experience) and oftentimes involves rewriting entire sections of code.

Despite all of this, AI is certainly a very useful tool. To create a minimum viable product, not quite ready for a product but nearly there, AI can help tremendously. Although it might not be the best at coding, using AI gives one access to vast quantities of knowledge to train. Debugging is also much easier with AI since it can understand error messages that are hard to find. AI can help a lot with programming even though it can’t replace programmers just yet.