@onusoz · /2026/07/12· 01:54 PM View on Anyone else notice that any non-codex harness is more token-efficient than codex on the same task? @evalstate· Jul 12, 2026I'm running Terminal Bench 2.1 with fast-agent (this is a subset while I try and get things working). Take this as raw, unchecked evalstate is normally wrong data but if we're talking about burning Codex quotas this might explain why I don't feel it so much?Show more Image hidden
@onusoz · /2026/07/12· 01:11 PM View on I SEE BURNS... BURNS EVERYWHERE @paulg· Jul 11, 2026This is like someone who can bench press 30 pounds writing about what it takes to bench press 600.Show more
@onusoz · /2026/07/12· 11:44 AM View on Here is the conversation where I built it. You can just prompt things @ratatui_rs highly recommended for TUIs. It just looks nice github.com/dutifuldev/annā¦
@onusoz · /2026/07/12· 11:26 AM View on I vibed a TUI in just 13 codex messages, and it works great @evalstate· Jul 12, 2026AnnoTUI by @onusoz is next levelš„ Annotate assistant responses, return quotes and comments for your follow-up (video should explain). IMO one of the biggest productivity unlocks rn. Used here with fast-agent (I have it bound to CTRL+X a), easy to integrate anywhere.Show more
@onusoz · /2026/07/12· 05:35 AM View on No better feeling than waking up to the smell of 5 successfully finished agents
@onusoz · /2026/07/12· 05:32 AM View on protip: use fable to write your READMEs and other user facing docs thank god fable exists, now that gpt 4.5 is gone