• Strategic programming is hard to learn because feedback often arrives months after the original decision. 1:01:12
  • Matt compares architecture choices to a mixing desk where you cannot hear the bad settings until much later. 1:01:38
  • AI shortens the feedback loop by allowing teams to move faster and create more code. 1:02:03
  • The codebase is the environment the agent operates in, so improving that environment improves agent performance. 1:02:27
  • Reading classic software design books gives engineers the language to reason about strategic choices. 1:02:49
  • The host notes that senior engineers often built judgment through years of painful production feedback. 1:03:21
  • Matt is unsure how companies should train juniors when tactical programming has become so cheap, but he says strategic knowledge is now more valuable because it has higher leverage. 1:04:26
  • To justify investment in fundamentals, Matt recommends measuring agent success and failure rates across the organization. 1:06:08
  • Agent observability is less invasive than developer observability because the organization is measuring a service it pays for. 1:06:30
  • Organizations should compare repos and teams, identify practices that improve agent success, and share those practices. 1:06:57
  • Matt argues teams need a shared set of skills and a common software workflow so they can experiment and improve together. 1:06:57
  • He recommends investing in the "factory" that builds software, meaning architecture, workflows, tests, and agent processes. 1:08:38