Agents are the users for MCPs and tools - so why not run "user tests" on them?

hot take: aws / azure / gcp UI is great …because i just use coding agent with CLI anyway and they know major cloud best—i’m not looking anyway

test driven development and functional programming officially won

it’s damn good

humans are the bootloader for the next mind and we’re already well into the terminal

codex is better

why are we doing last N messages for agents still? i barely remember the last 2 messages verbatim when texting

while back i had claude code get stuck on a bug and then rm -rf its own ~/.claude.json

so with these new smart models—does "stable authority assignment" create emergent responsibility-like behavior? 👀 Has anyone run another vending machine Anthropic