详细介绍Details
SpaceXAI 2026-09-22 官方文章:Cursor 并入后两支客服团队合并、产品线变多。做法是一步步放开:先接工单系统 Plain 和 Linear,让 Bot 像负责人一样处理工单,但只能写内部备注、每个写操作都要人批;给每次运行加记录和评估,出错就调;第一天先人工审它对简单工单的理解和回复,当天就放它直接回客户。之后让它在工单进来时先查 Linear 已知问题和 Datadog 报错、录视频复现 bug;按退款规则处理大部分退款;盯排队量和响应时限;同类问题多到阈值就自动开事故;每周给管理层报哪里答得不好。官方称工单量涨了 175% 没加人,单张成本低到 $0.20–0.30。 Official SpaceXAI post, Sep 22, 2026. After Cursor joined, two support teams merged around more products. Rollout was gradual: connect Plain (ticketing) and Linear, let the bot act as ticket owner but write only internal notes with every write approved, add traces and evals, review its read and proposed reply on easy tickets for one day, then let it answer customers directly. Later it pre-investigates every ticket against known Linear issues and Datadog errors, records bug repro videos, resolves most refunds under clear rules, manages queue priority and SLA alerts, declares incidents when similar reports cross a threshold, and sends leadership a weekly note on where AI answers fall short. SpaceXAI reports a 175% ticket increase with no new hires and costs as low as $0.20–0.30 per ticket.
为什么推荐Why we recommend it 官方 News(2026-09-22)。收录理由:和已收的 Support 指南互补,重点是「怎么一步步放权」——先只写内部备注、全程人批,再放开。边界:175%、$0.20–0.30 等数字是官方自报,本库无法核实;用到的 Plain、Linear、Datadog 需要你自己有。本库未实测。 Official News (Sep 22, 2026). Why listed: complements the Support guide already listed, focusing on how to hand over control step by step, starting with internal notes and human approval on every write. Boundary: the 175% and $0.20–0.30 figures are self-reported and not verified here; Plain, Linear, and Datadog are your own to connect. Not tested here.
上手路径How to start
审批边界Approval boundary 起步阶段所有写操作都人批;放开后直接回复和规则内退款由 Bot 做,规则外的仍交给人。 At the start every write is approved by a person; later the bot replies and handles in-policy refunds itself, and anything outside the rules goes to a person.
- 先跑一次安全范围Run a safe first pass
先在可撤回、不外发的范围内试跑。本条未批准前:起步阶段所有写操作都人批;放开后直接回复和规则内退款由 Bot 做,规则外的仍交给人。。 Try it first in a reversible, non-outbound scope. Until approved: At the start every write is approved by a person; later the bot replies and handles in-policy refunds itself, and anything outside the rules goes to a person..
- 存 SkillSave a Skill
把跑通的步骤存成 Skill,下次同类任务直接复用,而不是重新口述。 Save the working steps as a Skill so the next similar job reuses them instead of being re-explained.
- 再上 RoutineThen add a Routine
输出稳定后再打开例行 / 定时。本条已有例行形态,不要一上来就无人值守外发。 Turn on the routine / schedule only after output is stable. This item already has a routine shape — do not start unattended outbound.