Claire Vo on OpenClaw, multiple agents, and building a real AI assistant

Mar 29, 2026 · Source ↗

Published 2026-03-29 - Runtime about 107 min - Watch on YouTube

TLDR

Claire Vo says OpenClaw became useful only after she stopped forcing one agent to do everything and split work across multiple specialized agents.
The biggest unlock is practical automation: family scheduling, inbox triage, sales, and course management, with trust increased step by step.

Key Takeaways

Claire Vo moved from an eight-hour install that deleted her family calendar to running about nine OpenClaw agents across three computers.
She treats OpenClaw like an assistant with progressive trust: calendar first, then email reading, then drafting, then sending.
Telegram is her preferred setup channel, and she says voice-note onboarding through the “Yappers API” is high-bandwidth.
She recommends better models like Opus 46, Sonnet 46, and GPT54 for security hardening and a smoother experience.

Notes

Claire Vo hosts How I AI, runs ChatPRD, and says OpenClaw now supports her business, podcasting, and family life.
Her first OpenClaw install took eight hours and deleted her personal family calendar, which initially made her a skeptic.
She says the product’s real value appeared after repeated use, not on day one or even week one.
She now runs about eight to nine agents on OpenClaw across three Mac minis and other machines.
She names agents Polly, Finn, Max, Howie, Kelly, Holly, Sam, and others; Sam handles sales and replaced about 10 hours a week of paid work.
Her first default use case was a general assistant for scheduling, email, personal project management, and chief-of-staff tasks.
The strongest fit is household coordination: three kids, two schools, a baby, three basketball leagues, soccer, ballet, and mixed work and social commitments.
She says her husband is a major enabler because he shares house and professional responsibilities.
OpenClaw matters to her because it is open source, so users can inspect docs and code rather than trust a closed system blindly.
She warns about prompt injection and external instruction attacks, including malicious emails and poisoned websites that try to steal secrets.
Her safety approach is to keep the agent personal by default, restrict channels, and expand access gradually after trust is earned.
Setup starts on a separate computer, with Chrome, a separate Gmail account, and a local account; then you paste the install command from openclaw.ai.
Telegram was her beginner-friendly channel choice, though she notes it requires messaging Telegram’s botfather.