Legendary investor Naval: In the AI era, traditional software engineers have no value?

Original Title: Is Traditional Software Engineering Dead?

Original Author: Naval

Original Compiler: Ken, ChainCatcher

Is traditional software engineering dead?

"Does this mean that traditional software engineering has vanished? Absolutely not. Software engineers— even those who are not necessarily responsible for tuning or training AI models— are now among the most leveraged individuals in the world. Of course, those who are responsible for training and tuning models have even greater leverage, as they are building the toolsets that software engineers use.

But software engineers still have two huge advantages over you. First, they think in code, so they truly understand the underlying mechanisms at play. Moreover, all abstractions are flawed. Therefore, when computers program for you— when Claude Code or similar programs code for you— they are bound to make mistakes.

They will produce bugs, and the architecture may not be ideal. So it won’t be perfectly normal. Those who understand the underlying principles can plug the gaps when flaws arise.

So, if you want to build a well-architected application, or even just want to be able to accurately articulate the requirements for a well-architected application, if you want it to run at high performance, to operate at its best, and to catch bugs early, then you need to have a background in software engineering.

Traditional software engineers will be better positioned to leverage these tools. Furthermore, there are still many problems in software engineering today that these AI programs cannot handle. The simplest way to understand this is that these problems are outside their training data distribution.

For example, if they need to perform binary sorting or reverse a linked list, they have seen countless examples of such tasks, so they are very good at it. However, when you start to go beyond their domain— when you need to write extremely high-performance code, when the code needs to run on novel or entirely new architectures, when you are truly creating something new or solving new problems— you still need to dive deep and write the code yourself.

At least until there are enough examples accumulated to train new models, or until these models can perform higher-level abstract reasoning and tackle problems independently...

And remember: mediocrity has no market demand. No one wants a mediocre application unless it fills a niche that even better applications have failed to cover. Better applications will almost always capture 100% of the market share. There may be a small portion of users who shift to the second-best application because it performs better on some niche feature, or because it is cheaper, and so on.

But overall, people always want the best. So the bad news is that being second or third is meaningless— just like the famous line from Alec Baldwin in the movie Glengarry Glen Ross: 'First place gets a Cadillac Eldorado, second place gets a set of steak knives, third place is fired.'

In these winner-takes-all markets, this is absolutely true. The bad news is: if you want to win, you must be the best at something.

However, the things you can do to be 'the best' are endless. You can always find a perfect fit for your niche and become a leader in that field. This echoes a tweet I posted before: 'Strive to be the world’s best in your field. Constantly redefine what you do until that statement becomes a reality.'

I believe this still holds true in the age of artificial intelligence."