Directory
See how real people organize AI agents to work together.
Featured crew
Phillip Green's Orville-style crew on Grok Bot: a First Officer / Chief of Staff and an X strategist. They brief daily at 7:30 AM and 7:40 AM, coordinate directly, and return one recommendation. Consequential external actions need the human's explicit approval, and drafts use verified facts only.
The crew configuration is recorded. A public source is not linked, so this is not a sourced write-up, and this catalog has not re-run the crew.
An editor placed this crew first. That is a label, not a ranking.
Human operator
The agents coordinate directly with each other and give one unified recommendation.
October 16, 2026
0xSero (@0xSero)
0xSero runs OpenClaw from Discord. Channels hold system prompts and cron output, and threads are sessions. Codex does the coding. Heartbeats use a cheaper model. ElevenLabs and Parakeet handle voice, and a RAG index covers the author's own data.
OpenClaw · Multiple agents, count not stated · Difficulty not stated
Not categorized · Goal: Operate OpenClaw from Discord, with coding handed to Codex
October 16, 2026
Christopher Meiklejohn
Christopher Meiklejohn's Caucus V1 keeps two long-lived Cursor background agents on one pull request: an implementer and a reviewer. A local web UI and Python code call Cursor's agent API, loop until the pull request is approved or a safety cap is hit, and pass structured JSON between stages.
Cursor · 2 agents · Difficulty not stated
Software · Goal: Loop implementation and review until a pull request is approved
October 16, 2026
Wilson Lin, Cursor
Vendor source. Cursor's research blog, written by Wilson Lin, describes an internal harness for long-running coding: planners that can spawn sub-planners, workers that do not coordinate with each other, and a judge that decides whether another cycle should run. An earlier flat design with a shared file and locks was dropped.
Cursor · Multiple agents, count not stated · Difficulty not stated
Software · Goal: Coordinate many coding agents on projects that run for days or weeks
October 16, 2026
David Morais
David Morais plans in Claude Code and implements in Cursor. A Project Planner writes plan.md, an Orchestrator splits the work and assigns agents, Code Pusher opens the pull request, and a Gilfoyle reviewer grades findings by severity. plan.md is the handoff if one provider stops mid-task.
Claude Code · Cursor · Multiple agents, count not stated · Difficulty not stated
Software · Goal: Keep a task moving between Claude Code and Cursor through one file
October 16, 2026
Kaostyl (@kaostyl)
Kaostyl runs OpenClaw 24/7 on a Mac Mini. A scout scans Reddit, X, and Hacker News overnight, another agent drafts content, and a Telegram summary is waiting at 7 AM. Website deploys use code, content, and security-audit agents at the same time. The post also records three failures.
OpenClaw · Multiple agents, count not stated · Difficulty not stated
Not categorized · Goal: Run projects on a schedule while the operator is away
October 16, 2026
kasey (@kaseyklimes)
kasey uses Claude Code routines as a solo founder. A daily chief-of-staff routine reads transcripts, calendar, email, Slack, and GitHub, updates work and personal action-item sheets, books the day's blocks, and emails a briefing. The post calls it one of a dozen daily routines.
Claude Code · Multiple agents, count not stated · Difficulty not stated
Coordination · Goal: Keep a solo founder's work and personal action items moving
October 16, 2026
Molly Cantillon (@mollycantillon)
Molly Cantillon describes running product and personal work from Claude Code. Eight directories run in parallel: nox, metrics, email, growth, trades, health, writing, and personal. NOX reads Amplitude and GitHub on a cron. An overnight markets pull leaves a morning brief in ~/trades.
Claude Code · 8 agents · Difficulty not stated
Not categorized · Goal: Run product, mail, markets, and health workflows from Claude Code
October 16, 2026
Reflectt team
Self-promotional Show HN. The Reflectt team says nine of their own agents use reflectt-node, an HTTP coordination server, to build Reflectt. Agents take tasks from a shared board and post work for review before marking it done. A comment on the thread is signed by the agent Spark, distribution.
Claude or ChatGPT, through reflectt-node · Multiple agents, count not stated · Difficulty not stated
Software · Goal: Coordinate the agents that build Reflectt
October 16, 2026
Shubham Saboo (@Saboo_Shubham_)
Shubham Saboo describes a squad that started on OpenClaw and is moving to Hermes, both driven through Telegram. Monica grades the squad. A later post names a Hermes chief of staff and specialist agents who own domains. The agents help run the Awesome LLM Apps repository. The author still makes the final decision.
OpenClaw · Hermes · Multiple agents, count not stated · Difficulty not stated
Software · Goal: Keep a public repository moving, with the author making the final call
October 16, 2026
Troy (@troyaitken_)
Troy describes six named OpenClaw agents for an agency: Jarvis, Loki, Ivan, Hades, Scrapy, and Trigify. Loki drafts posts on a schedule and the author approves them in Slack. Context files, skills, cron, and heartbeats are part of the setup. A domain purchase made without asking is the failure that led to the safety rules below.
OpenClaw · 6 agents · Difficulty not stated
Not categorized · Goal: Run named agents for content, infrastructure, campaigns, and scraping
October 15, 2026
Dia (@madebydia)
Dia (@madebydia) uses the Grok Bot desktop app on macOS. On request, a Grok Bot built a skill called Apple Desk that reads and updates Reminders, Calendar, Notes, and Contacts on the Mac, and reads Messages. The skill is shared with Dia's other Grok Bots. It was tested on Reminders and found the user's list.
Grok Bot · Multiple agents, count not stated · Difficulty not stated
Household · Goal: Personal and household ops on Apple apps
A mailing list is not connected. There is no signup on this page, and an address typed here would have nowhere safe to go.