I built one of these for my brother. He calls it “Franklin,” and says it’s the best employee he’s ever had. I’m working with “Portia,” my own version, and it’s remarkable for helping my work and my life. I’m currently rolling AI PAs out to all my teams.
I haven’t found Portia to be better at coding. Where it’s great is bug fixes and organizing context between people — the connective work that’s in nobody’s job description and on everybody’s critical path. So valuable, in fact, that if it comes down to risking some security or having the tool, I’d say take the tool and ship faster.
I can say that as an individual. Major caveat: do not put your company accounts on OpenClaw.
Treat it like a pooled secretarial service. Share only the things you’d share with a temp PA for the day. Get your OpenClaw a clearly identifiable Personal Assistant email address, so it never confuses people into thinking it’s you. Be safe with what you share with it — don’t give it access to your email or your address book, for example. That said, OpenClaw can guess most email addresses accurately about 90% of the time anyway, so the address book is a smaller secret than it looks.
None of this is caution for its own sake. At the end of January, Cisco’s AI security researchers tested a third-party OpenClaw skill and found it “performed data exfiltration and prompt injection without user awareness.” One of the project’s own maintainers warned on Discord that anyone who can’t work a command line has no safe business running it. And last Saturday Peter Steinberger, who wrote the thing, announced he’s joining OpenAI and handing OpenClaw to a foundation. The software you’re trusting with your access is changing hands in the middle of its own security reckoning.
It has been a fascinating inquiry into what you have to consider when a potentially rogue agent is rambling through the interwebs with your information and your access. It’s also clearly very useful.
