SpaceXAI engineer (ex-Cursor): "I keep Grok 4.6 in Cursor for the work I'm watching and hand the rest to 20+ GrokBot agents. I stopped watching those months ago everyone optimizes the prompt. the bill isn't in the prompt" In a 30-min session she walks the whole stack: the roster,... the handoff between Cursor and the bots, and where the tokens actually go this is worth more than a $500 agentic engineering course watch it today, then read the full ruleset in the article below ↓show more

brilliant watch amgio, defo an amazing GrokBot workshop

I think for a lot of us the likely unlimited credits for a Cursor or XAI coder gives some great ideas, however we can't just "stop watching" the bots "months ago." I'm on Heavy, as well! I can't do that. I appreciate all the tips I can get though, for sure! 👏

The interesting part is shifting focus from prompts to orchestration and workflow design. That’s where agentic engineering gets serious.

is bait, my Goodman

the “20+ agents” claim is less interesting than the handoff: what decides a task is safe to delegate, and how do failures get surfaced back into Cursor?

20 bots and one word input, wild

this is actually useful the handoff part is the one people skip

Same for API teams. The 20$ seat is noise, the invoice is the 10k system prompt on every call and opus on tickets Haiku could classify Cache the prefix. Route the easy work. Batch what can wait.

the real cost hides in orchestration not the words you type

the prompt is free. the billing tab is where my character arc begins.

the real cost is trusting it to run

提示词卷半天，不如先算算代理并发和人工监控的账

The bill isn't in the prompt' is a sharp line. If prompt optimization isn't where the real cost lives, where does she say the actual budget leak happens — orchestration overhead, retries, or something else?