WHAT NEW IN X11.09 MON

AI collections from X · 25 collections · 74,882 posts

personal-agentVIDEO

Grok Bot: a 9-step setup for parallel multi-bot teams

A step-by-step guide to splitting Grok Bot into researcher, writer and reviewer roles, with handoff rules that let six bots run 2.4x faster than one.

rvaniaaa
@rvaniaaaaArchived 16 days agoOriginally posted 2026.10.23
rvaniaaa @rvaniaaaaOriginally posted 2026.10.23
Open original on X

Grok Bot teams are built for parallel work. Here is the exact setup that makes 6 bots run 2.4x faster than one. One line: researcher → writer → reviewer. Each waits for the one before, so six bots in a row run as fast as one. Parallel: inbox triage, calendar check and competitor page run at once, then one bot merges them into a morning brief. (2.4x is 6 bots with 70% of the work independent.) step 1 → Draw the arrows before you create a bot. For every "and then", ask: does this step need the last one's output? Yes is a real arrow. No means those two steps can run together. step 2 → Build around parallel work. Split the job into inputs that don't need each other, then give one bot the job of merging them. step 3 → One owner per stage. The docs say it themselves: "Ask for a single owner at each stage." Without one, parallel handoffs make duplicate work. step 4 → Every bot earns its place: its own schedule, a check of work it didn't do, narrow memory, or real parallel work. No reason, no bot. step 5 → The reviewer lives in its own conversation. It gets the artifact, the source and the criteria, never the thread. Fresh eyes every time. step 6 → Pass paths, not piles. Each bot saves a 15-line summary to a shared file and hands off with one message: task id, file path, what it needs back. step 7 → Cap every chain. 6 handoffs per task. Stop after 2 rounds in a row with no progress. Write both rules into every bot's description. step 8 → Log every handoff to one file and run a 43-line script on a schedule that flags runaway chains. A loop can't hide anymore. step 9 → Plan the night. Background approvals expire after about 10 minutes, so every overnight chain ends in an action you allowed in advance or a draft saved in a file. You review in the morning. The team never stalls. step 10 → Audit the whole team from a model outside it. It marks the fake arrows and the bots that should merge. One check, and the team is tighter. Same bots. Different arrows.

This post quotes another post: x.com/rvaniaaaa/status/2112997845854797838

x.com/rvaniaaaa/status/2113580278882996551Archived 2026.10.24 10:38