• Mantinband says agents are extremely good at writing code in any language, including unfamiliar or newly invented languages when given a clear brief. 1:49
  • He says recent models have changed his view: over the past several months, Opus 4.6 and 4.7 have produced production-ready output on complex codebases for him. 1:49
  • He argues it is naive to assume agents will not outperform human developers in output quality. 2:27
  • Agents are also strong at taking a direction, planning the implementation in detail, and then executing it. 2:27
  • His practical conclusion is that writing code by hand is no longer a good use of developer time when an agent can produce the code and the human can review or guide it. 2:27 2:58
  • Agents work especially well inside known frameworks and solved problem spaces. 5:45
  • He believes agents are approaching the point where reading every generated line of code will become obsolete, with tests and high-level validation replacing line-by-line review. 6:19 6:54