Coloured workflows.
We have coloured functions, because the async spreads.
Today, the AI spreads. When you put an AI workflow in contact with a human one, the humans must yield. It’s a fire hose that can only be consumed by another fire hose pointing yet further downstream.
@thdxr@rekram11 hey guys, found another one, `opencode` 1.0.203 -> /connect -> anthropic -> api key -> paste `sk-ant-oat01-gbd1xIKg6w3d...` -> bun crash https://t.co/uSPTgFR20p can you repro?
@elithrar@rekram11@thdxr yeah, seems like its ignoring the built-in agents as soon as you add one custom agent
(note: im on a dev vm, and claude is my sysadmin, and its install a BUNCH of stuff in whacky ways; this said, id hope that opencode could be resilient to this...)
@thdxr was there a regression in opencode? im suddenly seeing `^[\Configuration is invalid: No primary agents are available. Please configure at least one agent with mode 'primary' or 'all'.` i have 1 sub-agent in the ~/.config/opencode/agent/ directory
@yacineMTB True, because of the AI tide. Level is junior-to-mid right now, if you’re above it, you can ride it; parallelism increases with your height. But below it, you drown after 2 prompts
Doctors: we've accepted too many myths about Covid.
My presentation explains why - and gives you the science you're missing.
https://t.co/YxBxZwW18M
🧵 1/11
@nilotpaul_n I discovered this about 6 years ago, did it for 4 years and then went back to inline handling. It's cleaner and simpler. The trick is more helper functions: x.JSON(w, <any>) and x.JSONErr(w, 400, "my error: %s", err). Test with httptest.
@aramh Tiny modules are bad because they force you down a road of many tiny black box integrations. These interfaces come with a steep complexity cost.
It’s better to have larger black boxes (Go package directories), and right-size them for human brains.