Recently, in conversations around tech hiring and technical discussions, I’ve noticed a recurring question: “How do you use AI in your daily workflow?”. Every time it comes up, it feels a bit strange.

To me, a more revealing question would be: “In which situations do you find AI completely useless?”. If a person truly understands AI, they should know exactly when not to use it.

The current world

It has been a while since OpenAI dropped the first version of ChatGPT and Anthropic released the Claude Code CLI agent in February 2025. Yet, reality hasn’t quite matched the early predictions. The belief that AI will automatically boost productivity remains a core gospel in the tech world. Daily, when I open social media platforms like Facebook, X, or LinkedIn, my newsfeed is flooded with posts about AI tech. Almost all of them follow the exact same pattern:

“It used to take $X$ time to do $Y$. With AI, it takes only $0.1X$ to do $Y$—or in $X$ time, we can now do $10Y$.”

This framing pushes the assumption that simply increasing output ($Y$) equals better results. The focus becomes entirely on maximizing quantity.

The relationship between volume and value

A direct relationship between volume and value makes sense in physical manufacturing - example as paperclip factory. When every individual paperclip sold contributes directly to revenue, increasing production volume yields higher business value. Naturally, factories introduce machines to maximize that volume.

Applying that same logic to software engineering leads to the expectation that AI should primarily drive an increase in lines of code and feature counts. But a software feature is not a physical paperclip.

Ten paperclips provide identical functionality. In software, 10 features must solve 10 distinct problems.

Software problems are abstract and uncoupled from physical constraints. Because each feature addresses a different problem, blindly adding features doesn’t automatically help a user solve their primary challenge.

When applied to software, the statement that increasing volume (features) creates higher value start to falls apart.

AI should be used to build a good solution, not maximum amount of solutions

The ultimate goal for any company is to survive, which requires generating revenue by selling a good product, not a product bloated with endless features.

The narrative circulating today implies that using AI will automatically yield good products. That simply isn’t true. Great software was built long before AI existed; Linux, Python, and foundational platforms were crafted by sharp human minds.

Key takeaway: A good product is built by skilled people using AI; it is not the automatic result of assuming “using AI produces a good product.”

Personal skill and domain knowledge remain the core elements of development. Using AI helps build a good product faster, but it does not replace the craft itself.

AI is simply a tool for problem solvers

I view AI as a multi-purpose tool. I treat it like an IDE; models from OpenAI, Anthropic, or Qwen are conceptually similar to tools like VS Code, Vim, or Sublime. If I want speed, I can use AI to write code. But when I want to deeply implement my thinking, I craft the code myself.

Similarly, it could be a non-biological partner. I might think through a solution on my own, or use AI to explore alternative options I hadn’t considered.

The mark of a skilled engineer isn’t how much AI they use, but knowing when to put it down. It doesn’t matter which IDE you use to write your code; as long as the code satisfies the compiler or interpreter and solves the actual problem, it works.