I hate that this is true

TL;DR

• 1

  AI raises the floor for bad engineers: Theo agrees with Sean Goedecke's claim that tools like Claude Code turn truly chaotic pull requests into something at least line-by-line functional, even if it is still weird or wrong in deeper ways.

• 2

  Team quality gaps matter more than headcount: He ties this to The Mythical Man-Month and says adding engineers often slows projects down, which is why small, elite teams are still winning at companies like Twitch and in his own work on T3 Chat.

• 3

  His Twitch blog horror story is the whole thesis in one example: A principal engineer turned simple MP4 demos into 80 MB to 200 MB GIFs, then broken Twitch embeds, then literal pasted HTML text, costing Theo weeks of time, a near firing, and possibly a promotion.

• 4

  The big distinction is motivation, not just skill: A junior engineer like Mel, who built complex systems before even knowing what an endpoint was, can use AI as an infinite tutor and ramp absurdly fast, while lazy engineers use it to hide stagnation.

• 5

  Strong engineers still get more out of AI than weak ones: Theo says the best people already have taste and can catch obvious bad output, so as models improve, good engineers will compound faster and the gap with the bottom 30 percent will widen.

• 6

  People with rigid bad opinions can still sabotage AI: He calls out the anti-React archetype as a case where engineers force agents into worse technical decisions, proving AI helps only when the human guiding it is open to learning.