Skip to main content

[@DwarkeshPatel] The Three Types of Programmers in 2026 - Andrej Karpathy

· 2 min read
[@DwarkeshPatel] Summarizer
Video Bot

@DwarkeshPatel - "The Three Types of Programmers in 2026 - Andrej Karpathy"

Link: https://youtu.be/MqarkPmYI9Y

Duration: 0 min

Short Summary​

People interact with code through three major classes: those building from scratch, intermediate developers serving as architects, and agents executing tasks via prompts. Intermediate developers leverage LLM models to write code and utilize autocomplete functionalities. Additionally, agents excel at handling boilerplate tasks and copy-paste operations supported by extensive training data.

Key Quotes​

Key Quotes​

  1. "I would say there's like three major classes of how people interact with code right now. Some people completely reject all of LLMs and they are just uh writing by scratch. The intermediate part which is where I am is you still write a lot of things from scratch but you use the autocomplete uh that's basically available now from these models but you're still very much the architect of what you're writing. And then there's the you know vi coding hi please implement this or that uh you know enter and then let the model do it and that's the agents." (00:00:00)
  2. "I do feel like the agents work in very specific settings." (00:00:29)
  3. "They're very good at stuff that occurs very often in the internet because there's lots of examples of it in the training sets of these models." (00:00:35)

Detailed Summary​

Code Interaction Framework​

  • People engage with code through three distinct classes: those rejecting LLMs for manual writing, intermediate developers acting as architects, and agents executing tasks via specific prompts.
  • Intermediate developers maintain a crucial architectural role by using available LLM models to write code from scratch alongside standard autocomplete functionalities.
  • Agents operate effectively in specific settings where users provide targeted prompts to implement tasks and allow models to execute them autonomously.
  • LLM agents excel at handling repetitive boilerplate tasks and copy-paste operations, leveraging their extensive training on frequently occurring internet examples.

Operational Outcomes​

  • The integration of these classes ensures that boilerplate tasks are managed efficiently, reducing manual coding overhead.
  • Training sets play a vital role in enhancing agent performance for copy-paste operations and complex task implementations.
  • The collaboration between developers and LLMs fosters a dynamic coding environment capable of adapting to diverse project requirements.