The video explores the evolution of programming, arguing that we are currently witnessing a fundamental shift in the nature of the profession. Historically, programming has been the act of translating vague human desires into precise machine instructions. While various tools (Fortran, COBOL, No-Code) have repeatedly “predicted” the end of programmers, they actually increased software demand (Jevons Paradox). However, the emergence of Large Language Models (LLMs) is different because the machine is now learning human language. The thesis is that while “programming” as a skill won’t die, the role of the “coder” (one who writes lines of code) is being replaced by the “architect” or “decider,” and eventually, the “verifier.”
Historic Context
- The Dawn of Computing (1946): The ENIAC, a 30-ton machine, was programmed by six women who manually wired and flipped switches. These women were the first programmers but were largely ignored in the public celebration.
- The Evolution of Languages: From manual wiring to typed codes, and later to high-level languages like Fortran (1954) and COBOL, which aimed to make programming more accessible to non-experts.
- Human Computers: Before electronic computers, “Computer” was a job title for people (mostly women) performing manual calculations for institutions like Harvard Observatory and NASA.
- The Transition: The last generation of human computers (e.g., Dorothy Vaughn) became the first generation of electronic programmers by learning languages like FORTRAN.
Facts and Persons
- Key Figures:
- Henrietta Levitt: A human computer at Harvard who discovered patterns in pulsing stars, enabling the measurement of galactic distances.
- Catherine Johnson: A NASA human computer whose manual verification of trajectories was trusted by astronaut John Glenn over the electronic computer.
- Dorothy Vaughn: A leader at NASA who anticipated the shift to electronic computing and taught herself and her team FORTRAN.
- William Stanley Jevons: Economist who described the paradox where increased efficiency of a resource (coal) leads to increased total consumption.
- Edsger Dijkstra: Computer scientist who argued that natural language programming was impossible due to inherent ambiguity.
- Andre Karpathy: AI researcher who noted that “English is the hottest new programming language.”
- Technical Facts:
- ENIAC: Contained nearly 18,000 vacuum tubes.
- AI Adoption: Google reported that AI wrote 25% of new code in 2024, 50% in 2025, and 75% by April 2026.
Technical and Science Discoveries
- From Precision to Guessing: Traditional programming required absolute precision. LLMs shift this by becoming “good enough” at guessing human intent from ambiguous natural language.
- AI Agents: The move from simple code suggestion to agents that can plan, write, test, and self-correct across entire projects.
- Feedback Loops: The discovery that AI models trained on AI-generated data can degrade (the “photocopy of a photocopy” effect), highlighting the critical need for human-generated data.
Consequences
- Professional Shift: A decline in traditional “programmers” (those focusing on the act of coding) and a stability or shift toward “software developers” (those focusing on what to build).
- The “Magic” Civilization: The risk of creating a world run by AI-generated software that no living human can read, verify, or truly understand—effectively turning technology into “magic.”
- The Role of the Verifier: The emergence of a new critical role: the human who understands the underlying logic well enough to check the machine’s work, mirroring Catherine Johnson’s role for John Glenn.
- Intellectual Value: The argument that learning to program remains essential not for job security, but to understand the logic of the world and avoid blind reliance on machines.
