Agency over intelligence, and 2025
Andrej tweets a lot of useful insights, but this one hits differently. Intelligence and experience will give you an upper hand, but only if you’re willing to invalidate every aspect of how you’ve worked over your career (see Boris, inventor of Claude Code, in the replies). Even now I see some engineers (usually senior) taking jabs at the AI integrations in our workflows at Google. Or they push the narrative from upper management, but you can tell they haven’t extensively tried the very things they’re pushing. I have no reservations about using agents to write my CLs. In fact, one key benefit as Gemini reaches parity with other frontier models is that I have unlimited quota.
I know the job market for new engineers is tough, but sooner or later even big institutions will realize that having this “AI-native” fresh blood is hugely beneficial for adopting new tools.
2026 is shaping up to be an incredibly exciting(?) year in so many ways.
Only a few days left.
Below are some things I enjoyed listening to and reading this year:
Shin Ji Hoon: memories, soon to be a chapter of an anthology
Dwarkesh Podcast
Works in Progress magazine
Some tools I replaced:
Raycast instead of Spotlight
Helium, the browser (RIP Arc)
Orbstack instead of Docker
Gemini instead of ChatGPT for general queries (have you tried the live mode?)
Claude models instead of GPT-* models for coding
Ghostty instead of iTerm2


