Re: The End of Software

🌈 Abstract

The article discusses the potential impact of AI, particularly Large Language Models (LLMs), on the software engineering industry. It argues against the notion that AI will lead to the "end of software" or the replacement of software engineering jobs.

🙋 Q&A

[01] Why AI Won't Kill Software Engineering Jobs

1. What are the key points made in the article about why AI won't replace software engineering jobs?

• The article argues that just because the cost of creating software may decrease due to AI, it doesn't necessarily mean bad things for the software industry. It draws a parallel to the computer industry, where prices have dropped significantly, but the industry has thrived and an entire ecosystem of developers and support has emerged.
• The author expresses doubts about the ability of current LLMs to fully replace programmers. The author's own experience with using ChatGPT to rewrite a Dart XML parser in Go suggests that LLMs can make significant mistakes that require manual fixes, and the author questions whether using an LLM would be faster than writing the code themselves.
• The author argues that LLMs are limited in their capabilities because they only predict the next word, rather than thinking like a human. This means their output will always contain "hallucinations" and silly mistakes.
• The author cites examples of previous predictions about the "end of programming" that did not come to fruition, and suggests that people may be overly worried about programming jobs due to the current economic climate and layoffs.
• The author argues that software is too important and complicated to be fully outsourced to AI, and that programming jobs are unlikely to go away, even if some jobs like artist roles in the video game sector are being replaced by AI.

2. What evidence does the article provide to support the claim that AI won't replace software engineering jobs?

• The author's personal experience with using ChatGPT to rewrite a Dart XML parser in Go, where the LLM made significant mistakes that required manual fixes.
• The author's argument that LLMs are limited in their capabilities because they only predict the next word, rather than thinking like a human.
• The author's reference to previous predictions about the "end of programming" that did not come to fruition.
• The author's assertion that software is too important and complicated to be fully outsourced to AI.

[02] No-Code App Development

1. What are the author's views on no-code app development?

• The author initially had the view that "no-code is a scam", but has since changed their mind.
• The author acknowledges that no-code can be useful for building basic CRUD (create, read, update, delete) apps, as it can save a lot of money.
• However, the author argues that if you want to do anything more complex than a basic CRUD app, no-code becomes problematic, as you often reach the limitations of the no-code platform and have to export the code, which is then of poor quality and requires hiring a programmer to rewrite it from scratch.
• The author believes that no-code tools have actually increased the demand for programmers, as people who start with no-code often end up hiring programmers when their needs become more complex.

2. How does the author's view on no-code app development relate to their views on the impact of AI on software engineering jobs?

• The author's skepticism about the ability of no-code tools to fully replace the need for programmers mirrors their skepticism about the ability of LLMs to fully replace software engineers.
• Just as the author argues that no-code tools have their limitations and often lead to the need for programmers, the author also argues that LLMs have limitations and are unlikely to completely replace the need for human software engineers.