Because of AI, everyone needs to know programming languages

---

Jesen Huang in 2025

Jensen Huang famously said that with modern AI, the programming language becomes human language. Indeed, what used to take days to craft a program, AI tools can now write in seconds. Reports from the industry indicate that many senior engineers have not written a single line of code in months. However, AI coding is directed and managed by humans, and AI-generated code must be approved by humans. Code is cheap, but good code is not. In fact, more AI-generated code often means more human work. A compelling real-world example is the recent Anthropic source code leak, which highlights the risks associated with AI-assisted coding.

Fred Brooks in 1986

In his seminal article that laid the foundation for the field of software engineering, Frederick Brooks identified four essential difficulties of software development. He evaluated the solutions available at the time and concluded that while they addressed accidental difficulties, no solution could resolve the essential ones. He titled his article No Silver Bullet. The introduction begins:

Of all the monsters who fill the nightmares of our folklore, none terrify more than werewolves, because they transform unexpectedly from the familiar into horrors. For these, one seeks bullets of silver that can magically lay them to rest.

I imagine that Anthropic was probably horrified when their AI-generated system leaked over half a million lines of internal source code—a transformation from familiar tool to unexpected nightmare. Brook’s article is the required reading for the first week of the class I have been teaching. The first homework asks students to imagine talking to Jensen Huang and answering his question: “So, what are the essential difficulties of software?”

Programs Require Human Approval

Michael Scott, my colleague and the author of the popular textbook Programming Language Pragmatics, often quotes Donald Knuth: “Programming is the art of telling another human being what one wants the computer to do.” Part of this insight is that a program is a specification for a machine. In this regard, the key attribute is that a program must be precise. We should not—and must not—allow unspecified behavior by a machine. Think about programs that run laser eye surgeries, operate radiation machines, land jetliners, and control the ignition sequence to send Artemis II to the moon. We must write programs that are complete and precise.

AI has an unprecedented ability to automate programming, and its capabilities continue to improve. However, humans must be there to inspect and approve the generated code. The KPMG survey in Q3 2025 found that 63% of 130 executives said they are “putting humans in the loop due to a lack of trust, up from 45% last quarter.” In discussing the survey, one commenter explained that if people do not fully understand what a generated system does and why, they cannot defend it—not to a regulator, not to a client, and not to their leadership.

Precision is the foundation of reasoning. In mathematics, complete precision allows us to derive a result across a hundred steps while still knowing exactly what we are doing—and it enables someone to check every step along the way, a task that machines can now perform. Precision is the foundation of modern science and engineering. We can build complex systems today because we make every component fully precise.

The Conclusion: Everyone Needs to Know Programming Languages

With AI, modern programming may consist of three steps: (1) prompt AI to write code, (2) ensure the code is correct, and (3) explain to another human being what the code does and why it is correct. If Jensen Huang is correct that everyone will program using natural language, then it actually requires that everyone must know programming languages—because step (1) is fraught with risks without steps (2) and (3). Everyone needs to know programming languages enough to understand the code that they write using AI.