• Mantinband says developers should move up the abstraction stack: from machine language, assembly, C, and .NET toward specifying intent in English for agents. 12:59
  • His abstraction argument is that once a lower-level transformation becomes dependable, developers no longer need to inspect every lower-level detail, similar to not checking how the C# compiler lowers C# code. 13:32
  • He says complex features can require enormous shared context: in one project, he spent about 500,000 tokens just getting an agent aligned with his perspective and desired direction. 12:32
  • If an agent has a 1 million token context window and 500,000 tokens are spent on orientation, only about 500,000 tokens remain for the work, with quality decreasing as the conversation continues. 12:59
  • The recommended learning areas are software architecture, how systems interact, and cloud architecture. 14:05
  • Low-level details still matter for people whose jobs require them, such as developers working on compilers, low-level systems, or .NET libraries. 14:05
  • His practical advice is to learn .NET internals only when they are needed for the job, not in advance as a general-purpose career investment. 14:39
  • He sees companies that cannot use LLMs freely with their codebases as disadvantaged because agent-enabled companies can move faster. 14:39 15:10