OpenAI Dots works better split into four roles than as one big agent
It splits work across /digest, /lookout, /maker, /skeptic and /runner, and gives error-rate data comparing 5-step and 10-step chains.
Sam Altman OpenAI Dots gets way more useful when you stop treating it like one giant agent /digest → /lookout scans 3 sources: email + calendar + repo → /maker turns them into 3 buckets: SHIPPED / MOVED / WAITING → /skeptic checks the chain after 5 steps → /runner ships only after PASS the numbers are the interesting part: 5-step chains: 8.6% flagged 10-step chains: 19.7% flagged 2 shorter legs with checkpoints: 3.4% 4 seats, 1 shared memory, max 5 steps, under 5 approvals/day.
This post quotes another post: x.com/fleyta88/status/2105656875387625507