I used to have a habit of automating everything whenever possible. For example, when setting up shared initial configurations on Linux systems, I always wanted to complete the environment setup quickly after an instance was created, so I wrote some Bash scripts to help streamline the process.

However, I later realized that real-world tasks often involve different Linux distributions, and even when using the same distribution, different versions may have variations. Cloud service providers also introduce their own adjustments. In practice, I’m usually not the first person to interact with the system, and I often don’t have full visibility into any special configurations or unique requirements. This makes automation quite risky. Additionally, this kind of setup work typically only takes up a very small portion at the beginning of a product’s lifecycle, making full automation not necessarily worth the effort.

Even if AI eventually becomes capable of configuring and deploying systems on its own (which isn’t really possible yet), it would likely face the same challenges. In short, striving for 100% automation isn’t always the right approach.

Factors like accountability, understanding company culture, meeting customer requirements, and adapting to specific work habits are things machines simply can’t handle.

Humans have responsibilities—if something goes wrong due to negligence or recklessness, there could be legal consequences, loss of professional credibility, or even damage to their reputation. While humans may make mistakes and aren’t as perfect as robots, there are also irreplaceable capabilities that machines will never be able to replicate.