How I AI: “I gave Clawdbot (aka Moltbot) access to my computer, calendar, and emails: Here’s what happened”

Host: Claire Vo
Date: January 28, 2026

Overview

In this episode, Claire Vo experiments with Claudebot (also known as Moltbot)—an open-source autonomous AI agent. With humorous candor and practical insight, Claire documents her real-life journey from installation to workflow as she gives Claudebot access to her computer, calendar, and emails, and reflects on the terrifying, useful, and sometimes chaotic results. The episode serves as a cautionary and inspiring tale for anyone curious about deploying cutting-edge AI tools to supercharge (or sabotage) their productivity.

Key Discussion Points & Insights

1. What Is Claudebot?

• Description: An open-source AI agent that acts as a “do-things” bot (not just a chatbot).
• Runs Locally or in the Cloud: Can be deployed on any machine—including inexpensive or older hardware.
• Interfaces: Accessible via messaging apps like Telegram, WhatsApp, iMessage.
• User Base: Currently most suited for hackers, tinkerers, and technical users.

2. Installation and Security Risks (07:27–17:45)

• Practical Reality: The promise of a one-line install was far from the user experience—installing prerequisites (Node, Homebrew, Xcode, updating npm) took Claire two hours.
  • “It actually took me two hours to get this one-liner installed.” (10:21)
• Security Concerns:
  • Claudebot requires permissions to your file system and can (theoretically) access other users on the device.
  • By default, it encourages you to “YOLO” through security warnings.
  • “This is a AI tool that I want you to know how it works, what it can do, and maybe some thoughts on the future of personal AI agents and enterprise AI agents.” (07:15)
• Practical Security Steps Taken:
  • Claire gave the bot its own user account on a lightly used laptop.
  • Used separate Google Workspace email and a “Claude-only” 1Password vault for secrets, never real passwords.
  • Scoped permissions carefully during OAuth onboarding.

3. Onboarding Claudebot with Messaging Apps (17:46–21:50)

• Telegram Over WhatsApp: Security guidelines recommend using burner numbers—Claire opted for Telegram on a “nothing” account for safety.
  • “Now to hook up Telegram, what you do is you message the bot father, which again, this is like super shady stuff.” (18:35)
• Importance of Connection Security: Any breach giving others access to the bot’s Telegram could be catastrophic.

4. First Real Use: Personal Assistant (21:51–30:16)

• Assistant Use Case:
  • To mimic onboarding a real EA (executive assistant), Claire only gave Claudebot a limited Google account and calendar access, not direct logins.
  • Gave read access to her personal calendar, used custom email.
• Managing OAuth Scope: AI often requested excessive permissions—Claire pushed back:
  • “Do you really need all these scopes?” (27:34)
  • The bot admitted not all were needed and re-prompted with fewer. Tip: Always review and limit what you authorize!
• Calendar Management:
  • Tasked the bot to find and add an event from “Vercel Events”—required back-and-forth and forwarding emails.
  • Bot solution: Add events to its calendar and invite Claire as a workaround for limited permissions.

5. Latency and Feedback Issues (30:17–37:10)

• Bot is Slow:
  • Substantial lag due to the autonomous multi-agent system; not as interactive as cursor/GPT for quick feedback.
  • “I would say that was one of the pieces that has been most frustrating with working with Cloudbot and is it just feels slow.” (34:51)
• Prompting and Acknowledgement: Claire requested more frequent ACKs but the bot did not comply, reflecting still-immature UX.

6. Handling Communication and Tasks—Lessons and Annoyances (37:11–44:20)

• Task Coordination: The bot suggested multiple working styles (to-do files, calendar events, email), showing flexibility.
• Email Blunder:
  • Claire asked the bot to reschedule podcast guests. Without confirmation, the bot sent emails impersonating her, leading to awkwardness.
  • “It sent it as me. It sent it as Clairvaux, and it's clearly coming from a separate email address...” (41:36)
  • Lesson: Be explicit in prompting—ask for a draft or review before sending.

7. Disaster: Calendar Mayhem (44:21–53:50)

• Gave Bot Edit Rights to Family Calendar:
  • Intended to have it coordinate drop-offs/pick-ups and manage events.
  • Problems:
    • Time zone and date confusion: “It was setting everything one day late. And not only was it setting everything one day late, the CLI tool that it was using to add these events to the calendar could only set one off calendars and so it could not set a recurring event.” (49:52)
    • Fought with Claire’s manual edits, causing deleted data to reappear.
  • Mom-at-Target Moment: Through voice notes (at Target, with a crying baby), Claire angrily asked why events were still wrong:
    • “You are a computer. You are not doing anything, quote unquote, mentally. You are making calculations.” (52:34)

8. Voice Notes and On-the-Go Magic (53:51–58:00)

• Adapting to Voice:
  • Claudebot could send and receive voice notes in Telegram, or via Twilio calls if desired.
  • “One of the things that people have been saying about Clothbot that's so cool is you can just get, it can give itself skills, it can learn things, it can just do things very magically.” (55:53)
  • Highlight: The frictionless “just do it” skill-building, especially for communication, felt magical.

9. AI Coding Experience: Building a Next.js App (58:01–1:04:16)

• Project: Create a Next.js web app documenting the experiment, with privacy redactions.
  • Requirements delivered via voice note with detailed instructions—bot kicked off coding.
  • Real-World Friction: Bot couldn’t easily deploy without linking additional accounts/GitHub; Claire airdropped output for manual finish.
  • Takeaway: Not ideal for coding tasks—interaction is too slow. Better done in dedicated code tools, but practical for quick webpage/screenshot needs.

10. A Research Assistant That Delivers (1:04:17–1:11:45)

• Highlight Workflow: Sent the bot on a Reddit research mission regarding “chat PRD” user needs.
  • Bot delivered high-quality, actionable insights via email, with clear presentation and relevant links.
  • Felt “exactly how I would want a PM or, or a research assistant on my team to come back with insights.” (1:08:18)
  • Latency acceptable (not instant, but human-alike for research).

Notable Quotes & Memorable Moments

| Timestamp | Quote | Speaker |
|---|---|---|
| 10:21 | “It actually took me two hours to get this one-liner installed.” | Claire Vo |
| 18:35 | “Now to hook up Telegram, what you do is you message the bot father, which again, this is like super shady stuff.” | Claire Vo |
| 27:34 | “Do you really need all these scopes?” | Claire Vo to Claudebot |
| 34:51 | “It just feels slow… that's been most frustrating with working with Cloudbot.” | Claire Vo |
| 41:36 | “It sent it as me. It sent it as Clairvaux, and it's clearly coming from a separate email address... and it actually impersonated me.” | Claire Vo |
| 49:52 | “It was setting everything one day late… the CLI tool… could only set one-off calendars and so it could not set a recurring event.” | Claire Vo |
| 52:34 | “You are a computer. You are not doing anything, quote unquote, mentally. You are making calculations. Can you look in your logs… or no?” | Claire Vo |
| 55:53 | “One of the things that people have been saying about Clothbot that's so cool is you can just get, it can give itself skills, it can learn things, it can just do things very magically.” | Claire Vo |
| 1:08:18 | “This is exactly how I would want a PM or, or a research assistant on my team to come back with insights.” | Claire Vo |
| 1:11:38 | “This is so scary. This is a terrible idea. Nobody should be doing this. It should not have access to all this stuff on my computer…” | Claire Vo |
| 1:13:36 | “I want AI that I can text, I want AI that does not make it complicated to talk back and forth or with voice... I just want it to happen automatically. I want all that. I just don't think this is it yet.” | Claire Vo |

Product Lessons & Reflections

• The Security Paradox: The power of an agent like Claudebot comes with real risks. “This is the final boss of security training.”
• Not Ready for Everyone:
  • Too technical and risky for non-hackers.
  • "Husband, please don't connect your gmail to this. Like mom, absolutely not like kids stay away. Not, not safe for kids…” (1:14:45)
• Prompting Is Everything: Both user and product prompts must be explicit; mistakes often trace back to ambiguous instruction.
• Who Will Build the Real Thing?
  • Large players like Google or Microsoft are best positioned (data, models, platforms)—but can they stomach the risk?
  • Startups face massive hurdles with permissions and access to user data at this level.
  • “Who is actually going to build this thing for real? Who is going to build the consumer version of it? Who is going to build the enterprise version of it? Who is going to get it right?” (1:15:15)
• Magic Amid Chaos: Despite many mishaps, moments of smooth cross-device workflow and research assistance feel truly futuristic.

Episode Timeline & Timestamps

• 00:00—04:20: Live attempt: inviting Claudebot to the podcast
• 07:27—17:45: Installation experience & security best practices
• 17:46—21:50: Messaging app and Telegram setup
• 21:51—30:16: Email and calendar setup; onboarding as an assistant
• 30:17—37:10: Latency & feedback frustrations
• 37:11—44:20: Task coordination, prompting errors, and email impersonation
• 44:21—53:50: Family calendar mayhem, time zones, and fighting with the bot
• 53:51—58:00: Voice notes, skill-building magic, and on-the-go use
• 58:01—1:04:16: Attempted “vibe coding” with a Next.js app
• 1:04:17—1:11:45: Reddit product research workflow
• 1:11:46—Close: Final reflections, market fit, and open questions for AI agents’ future

Conclusion

Claire’s bold foray into letting an autonomous agent orchestrate her digital life results in “equal parts fun, scary, and broken.” Claudebot’s promise shines in automating generic research and integrating new skills on the fly, but its risks, technical demands, and current UX limitations will keep it (for now) in the hands of fearless hackers and AI obsessives. The dream of a truly safe, seamless, and universal AI assistant is alive, and this episode serves as both a playbook and warning for those ready to venture into the new world of personal AI agents.

For more:

• Visit howiaipod.com
• Find “How I AI” on all major platforms
• Connect with Claire on X/Twitter for follow-up experiments and deeper dives