Claire Vo says OpenClaw is the most important AI experience she has had since ChatGPT, after an eight-hour install that deleted her family calendar.
Vo runs nine OpenClaw agents across three computers, assigning each a role such as Sam for sales and Finn for family logistics.
Vo advises splitting tasks across separate agents on clean secondary machines, since shared machines let agents see each other's space and context overload hurts performance.
A leading OpenClaw skeptic explains how nine agents changed her life
OpenClaw rewards people who stay with it past a rough start. Claire Vo went from skeptic to running nine agents because she gave each one a narrow role, its own machine and clear instructions. She argues that managing agents is a management skill, so direction matters more than technical knowledge.
Takeaways
Judge tools by where they will be. Claire's first install took eight hours and ended with her family calendar deleted. She kept going because of the joy and utility, and now runs nine agents across three computers. She calls it her biggest AI experience since ChatGPT.
Split work across narrow, role-based agents. One agent given everything suffers context overload. Claire gives each agent a role, such as Polly for scheduling and Sam for sales, much like Slack channels. Agents sharing a machine see each other's space, so family agent Finn is moving.
Agents already take over routine paid work. Sam sweeps the ChatPRD CRM each morning for new signups and sends emails, replacing about ten hours a week Claire paid a friend for. Finn pings the family chat around three o'clock to sort school pickups.
Proactive behaviour comes from simple files and schedules. Onboarding has the agent ask who it is and who the user is, which fills identity files. A heartbeat makes it check its task list every 30 minutes, so Howie can send guest notes before each recording.
Use APIs over browsers and manage context. The open web is hostile to bots, so browser use is unreliable and Claire uses APIs first, including Exa for search. For memory she manages context and keeps a tools file rather than trying to harden memory.
Management skill matters more than technical depth. Claire credits her 20-plus years of management for her success with agents, since clear instructions drive results. Her agents assign her Linear tasks, and she uses Claude Code like a surgeon to repair OpenClaw configuration.
Setting up OpenClaw safely
Install it on a clean secondary machine, such as an old laptop or a Mac mini, rather than a personal computer.
Give the agent its own local admin account and email address, and treat it like a new employee who gains access step by step.
Restrict the agent to take instructions only from you on Telegram, which limits exposure to prompt injection.
Run the best available models, which resist prompt injection better and give a better experience.
Turn on screen sharing on a Mac mini so it can run without a monitor.
Tap a name to go there · tap a line to see why they are linked