Will AI replace programmers?

The recent launch of AlphaEvolve by Google DeepMind is truly amazing - it can automatically write and optimize algorithms and even make scientific discoveries. This can't help but make me think: is AI really going to replace programmers? As a practitioner who has been in the programming field for many years, my feelings are actually both excited and mixed. Judging from the trajectory of AlphaCode to AlphaEvolve, AI's coding capabilities are indeed advancing at an astonishing rate, but just how far can they go?

Let me start with an impressive statistic: AlphaEvolve's solution for optimizing Google's TPU matrix multiplication circuits saved 231 TP3T of energy compared to the solution given by a human expert. That's an impressive figure, but it also illustrates a key point - AI isn't about "replacing", it's about "optimizing". It's more like a super plug-in for programmers, allowing us to spend our time where creativity is needed. Think of all the repetitive code debugging time that could be saved!

Is AI a competitor or a partner?

When it comes to the current state of AI programming capabilities, I am reminded of a very interesting case. In a certain programming competition, the accuracy of the AI-generated solution was really high, but when it came to topics that required creative thinking, its performance began to be discounted. This shows that AIs currently have significant limitations in creative thinking. They are good at optimizing within known frameworks, but not at breaking the mold.

Take AlphaEvolve as an example, although it can be optimized on the basis of existing algorithms, the truly groundbreaking algorithm design still requires human intelligence. Doesn't this confirm the saying, "What AI replaces is not programmers, but programmers who won't use AI." It is likely that all future programmers will become AI curators and architects, focusing on defining problems, formulating strategies, and building system frameworks.

To put it another way, the essence of programming has never been about the code itself, but rather about solving problems, with AI handling the implementation details and humans controlling the direction and architecture - wouldn't that be a more reasonable division of labor? Just as a car driver does not need to know the principles of the internal combustion engine to drive a car, the programmer of the future may no longer need to master all the details of the underlying implementation.

AI brings about a change in the way programming is done

That said, the spread of technologies like AlphaEvolve will indeed fundamentally change the way programming works. It is expected that we may see a reorganization of the programming process in the next 3-5 years:

- Requirements Definition Becomes More Important - After All, AI Needs Clear Requirements Inputs
- Test-Driven Development Could Go Mainstream - Because AI Needs Clear Evaluation Criteria
- Programmers will spend more time on system design and code reviews

This is a natural evolutionary process, just as it was when stackers replaced manual compilation. The important thing is not to resist change, but to learn how to better navigate these new technological tools.

At the end of the day, I don't think AI will completely replace programmers, but it will completely reshape the industry. Those programmers who are able to live with AI tools will become stronger, while those who refuse to change may be eliminated. This is perhaps where the cruelty and charm of technological progress lies. After all, as a programmer, what excites me most is always the "problem solving" thing itself, not the way of knocking code.