Trading bot setup: Grok coordinates, Jev answers narrow questions
A 28-second animation shows a trading desk with 1 coordinator and 6 specialists: code computes prices and hard limits, Jev answers narrow questions, Grok handles task handoffs, and humans approve signatures.
1 coordinator, 6 specialists, and none of them gets to negotiate the loss limit... this 28-second grok desk animation puts research, synthesis, tools, planning, verification and memory around one coordinator. the bots show the handoffs. the pnl is simulated, not a verified trading result. the grok + jev article draws the more important boundary: which decisions need judgment, and which should never reach a model. → code calculates prices, exposure, position size and hard limits → jev answers narrow questions about the setup → grok coordinates tasks, reports and handoffs → humans approve signatures, fund movements and risk-limit changes a confident answer still has to pass the risk checks. missing data stays unknown. a late answer can mean hold. a rejected setup needs a recorded reason, not another prompt asking the model to reconsider. adding 6 bots doesn't automatically give you a trading desk. each role needs a defined job, a stopping condition and limits it cannot override. the moving characters make the coordination visible; trading history would have to establish whether it works. if a model can talk its way around your loss limit, you never had a loss limit.
This post quotes another post: x.com/0xkiyoro/status/2107135100911563017