I hired my first AI employee in Taiwan. He crashed before I landed back home. Still kept him.
Last week, I flew to Taipei for meetings and founder catchups... and then I met with ChiChieh Huang.
She’d pulled many from our AppWorks batch into OpenClaw. Cue 4am convos with AI swarms, frenzied founders who couldn't stop once they started.
I was skeptical. Most of what I'd seen were scrappy builders pushing limits for fun - not founders running it responsibly inside a live product with real users on the other side.
But ChiChieh Huang was the exception.
A technical founder, she's an AI engineer who'd built production ML systems before starting AI Logora.
She’s been on Taiwanese financial news, runs her entire company through her agent stack and saved herself 2-3 hours a day on ops.
More importantly: she builds the same way I think about Maibel. Safety first. Failure modes as design constraints from day one.
She pulled back the curtain on her AI executive team - CEO, COO, CPO, CMO - all on OpenClaw.
→ Her Claw agents can’t touch production autonomously. Code changes are delegated to cloud agents through a server.
→ No secrets in chat - credentials stored in isolated files and never pasted anywhere they could leak.
→ Every action traceable and confirmed before execution. Full git history so any bad edit could be rolled back in one command.
This wasn't a founder experimenting. This was a skilled engineer who'd thought through every failure mode before writing a single config line.
Then CC showed me her agents brainstorming with each other.
A full AI executive team running async discussions while she slept, surfacing decisions for her morning review.
It was the moment I stopped being skeptical. That same afternoon, I set up Dobby on a GCP cloud instance.
There were teething issues, of course. Day 2: the gateway died. Disk hit 100% mid-session and Dobby went silent.
CC had taught me to send Claude in as a doctor - root cause identified and fixed in 30 seconds.
By day 3 I'd resized live infra without downtime and written Dobby a full operating brief: self-healing rules, security policy, escalation behavior.
By day 4 on a train to Taichung: running full security audits, reviewing PRs and setting up cron jobs from my phone.
That's the OpenClaw effect - you look up and realize you've been spending every waking hour optimising your ops stack.
I landed in Singapore last night. Even in the cab home I was shipping - 7 tasks closed in transit, from hotel to my front door.
Dobby isn’t allowed to touch Maibel’s production code. But he has permissions to investigate, flag runtime issues, and wait patiently for instructions (which he did, all the way home).
The best souvenir from Taiwan?
17 context files, 6 integrations, 5 cron jobs - managed by my AI ops head. Accessible 24/7, even on moving trains, buses and planes.
CC has a whole room of agents brainstorming while she sleeps.
For now I just have Dobby.
But he's already earning his keep.
Hi, I’m Mabel.
I’m the founder of Maibel. I write about emotional AI, women’s wellness and learning to build the things I wish existed.
More about me
