Clawdbot has become very popular, and I really understand it. It's powerful. When you type a message in the chat box, it can do the work for you—emails, calendars, messages, automated processes—like a digital butler at your beck and call. Isn't that great? It is.

But let me say something harsh: this kind of thing, Clawdbot, is really not for most ordinary people. If you touch it, there's a high probability that something will go wrong, and you won't even know how you ended up in trouble.

What it wants is not "usage"; what it wants is "keys". If you want it to help you automatically handle emails, you need to grant it email permissions. If you want it to truly run, you have to stuff it with a bunch of tokens, API keys, and even higher system permissions. What is the most common mistake ordinary people make? To save trouble, they open all permissions. After you’ve opened everything, then talk to me about security; you’re just joking with yourself.

Ordinary people really don’t have that "risk intuition". You think you’re chatting with it, but in reality, you’re giving commands, and they are commands that can be executed. If you write something vaguely, it may misunderstand, and the execution could stray by a kilometer. Emails sent to the wrong person, schedules all changed, files deleted mistakenly, mass sending that leads to social embarrassment. You still want to recall it? Many actions can’t be undone.

The real pitfall is in "continuous operation". Once you hook it up and let it run automatically, trigger automatically, loop automatically, that’s no longer just occasional use. That’s a long-term online entity, constantly manipulating your account and your data.

If you misconfigure a rule today, you'll wake up tomorrow to find the entire workflow looks like it has been chewed by wild dogs. You ask who will take the blame? In the end, you will take the blame.

My attitude is very clear. Clawdbot is suitable for two types of people.

One type is very security-conscious, knows the principle of least privilege, knows how to isolate, and knows how to stop the bleeding when something goes wrong. The other type has a very narrow scope, only enabling a small set of functions, using it like a scalpel, not opening permissions recklessly, and not going full automation right away.

What is the most dangerous mentality for ordinary people? They think they’ve just installed a smarter assistant. As a result, they hand over a string of keys and still feel nothing is wrong. When something really goes wrong, you won’t even know "where the leak is".

If you really want to play with it, first remember one thing. Fewer permissions, slower functions, and more trouble are all better than a single disaster.