Clawdbot/Moltbot/Openclaw

Clawdbot/Moltbot/Openclaw

In the last week, Clawdbot quickly renaming to Moltbot now Openclaw, took the world by storm. It is an open-source project led by U.S. developers described as “The AI that actually does things” and do things it does, perhaps too many things…

The most comical proof of the hype is social media posts of people buying small stacks of Mac minis to run Openclaw around the clock, and there are plenty of claims that models have gone out of stock or slipped into backorder as a result of the craze.

Yet Openclaw is not quite the plug-and-play AI that some claim. It demands a degree of tech savviness the average user would not be familiar with. You are, in practice, signing up for command-line setup, fiddling with API keys, and maintaining a small web of tokens and credentials.

Nevertheless people are embedding Openclaw across their stack, then talking to it as though it were a colleague. Giving it access to their Whatsapp, messenger, emails, calendar and allowing it to run their life. Some go further, treating it as a back-office employee: managing invoices, tracking stock, even keeping the books for small business.

What I would say to this is: We’re firmly in the honeymoon phase. Early on, each user’s set-up is lean: a handful of “skills”, a manageable set of connections, a clear mental model of what the agent can touch. Over time, deprication, software updates, new skills and integrations accumulate, until the agent’s world is a tangled attic of tools and instructions. The more it can do, the harder it becomes for it to know what it should do. Something we know is not yet a solved problem.

Another important fact to consider is that we already know that large language models can be coerced, tricked, and socially engineered as they struggle to distinguish fact from fiction. Many implementations are one malicious prompt via pull request, metadata booby trap or document upload away from draining their owner’s bank account (if the API costs don’t bankrupt you first).

None of this means Openclaw is a dead end. It is, rather, a glimpse of what consumer agents might look like once security, scalability, maintainability and consistency has been solved. It brings the agent idea a step closer to everyday usefulness, even if it also brings the risks into sharper relief. For enterprise however, the road is longer: none of these shortcomings can be overlooked and it is unlikely the problems can be solved with engineering alone.