Author: Naval Ravikant
Compiled by: Felix, PANews
Amid the rapid iteration of AI large models, the global market is filled with deep pessimism and anxiety. OpenAI CEO Sam Altman predicted that "AI will take over 95% of programmers' jobs," while Anthropic CEO forecasted that "AI will fully take over software engineering roles within 6-12 months." The notion that "the programmer profession is dead" seems to have become a global consensus, facing the most severe "survival crisis" since the birth of the internet.
However, this fear of job disappearance stems from a misunderstanding of the underlying logic of technology. AngelList co-founder Naval Ravikant (early investor in Uber and Twitter) believes that the hype around AI's recent productivity boost may be overblown. No matter how advanced AI evolves, it will always make mistakes, and software engineers will remain an indispensable profession.
Regardless of the field you are in, even the smallest niche, as long as you master it and become a top talent, you need not worry about being replaced by AI.
Below are Naval Ravikant's latest views.
"Does AI mean traditional software engineering is dead?" Absolutely not. Software engineers—even those not necessarily responsible for tuning or training AI models—are now among the most valued people globally. Of course, those who train and tune models are even more valued because they build the toolkits used by software engineers.
But software engineers still have two major strengths. First, they think in code, so they truly understand the underlying mechanisms. And all abstractions are leaky. So, when a computer writes programs for you (e.g., using Claude Code or similar), it will always make mistakes.
It will produce bugs, have imperfect architectures, and generally not be entirely correct. Those who understand the underlying logic can patch the leaks when they appear.
Therefore, if you want to build a well-architected application, if you want the ability to define a good architecture, if you want your program to perform at high efficiency, achieve its best potential, and catch bugs early, you still need a software engineering background.
Traditional software engineers are better equipped to leverage these AI tools. Moreover, there are still many problems in software engineering that AI programs cannot solve. The simplest way to understand this is: these problems lie outside their data distribution.
For example, if you need to perform a binary sort or reverse a linked list, AI has seen countless cases, so it excels. But when you start venturing outside their familiar territory—such as writing extremely high-performance code, running on entirely new architectures, creating something novel, or solving new problems—you still need to manually write the code yourself.
This will continue until there are enough cases for new models to train on, or until these models can reason at a higher level of abstraction and independently solve difficult problems.
Remember: The market has no demand for 'mediocrity.' As long as a better application exists in a niche, no one wants mediocre ones. The better application will essentially capture 100% of the market share. Perhaps a tiny fraction will go to the second-best application, merely because it excels in some niche feature or is cheaper, and so on.
But overall, people only want the best. So the bad news is that competing for second or third place is pointless—like Alec Baldwin's famous line in the movie Glengarry Glen Ross: "First prize is a Cadillac Eldorado. Second prize is a set of steak knives. Third prize is you're fired."
In today's winner-take-all market, this is absolutely true. The bad news is: if you want to win, you must be the best at something.
However, the fields in which you can be the best are endless. You can always find a niche that suits you and become the best in it. This reminds me of a tweet I once posted: "Aspire to be the best in your field. Keep redefining what you do until it becomes true."
I believe this principle still holds in the AI era.
Related reading: A Memo from 2028: What Do We Lose If AI Wins?
