Grok Bot 工作坊：用 Jev 返回类型化路由决策
54 分钟讲怎么像产品团队一样管 agent，核心对比是前沿模型用散文输出路由，而 Jev 直接返回类型化选项和概率。
SpaceXAI just released a free Grok Bot workshop 54 minutes on running a bot team like an actual product org Grok Bot → Specialists → Shared Context → Delegation → Self-Improving Teams 7% → 03:34 - Why managing several agent threads gets tricky 25% → 13:30 - Organize agents around real company roles 44% → 23:58 - Have agents work on multiple questions in parallel 56% → 30:07 - Share context automatically, without copy-pasting 75% → 40:38 - See an agent learn to delegate and review tasks A frontier model returns every routing decision as prose Who takes the task → What needs checking → Who approves it That’s not an intelligence architecture Jev returns typed options and probabilities, without writing a sentence LLM creates the work → Jev picks the next step → Code enforces the decision Save this before building another coordinator agent ↓
这条是引用着另一条写的：x.com/0xwhrrari/status/2108905400653279271