Generative AI has done plenty of things in the three years or so since the release of ChatGPT. It has caused a massive stock market boom. It has integrated its way into our search engines and phone screens. And it’s become a tool that hundreds of millions of us use every day.
But despite all that, many people avoid using AI tools. Why? Because while tasking an AI with generating text, speech, images or video saves you time and effort, the process of getting it to do these jobs is clunky.
Typing the prompts required to get a chatbot to understand what you want it to do can quickly become a job in itself. And you’re still left with all the other annoying jobs that are part of modern daily life – replying to emails, booking appointments and paying bills.
That’s where the true promise of AI may lie – by taking care of those annoying mundane tasks on your behalf. So-called ‘agentic AI’ could do this and it’s perhaps what people are really hoping for from an AI – an always-on, eager assistant ready to carry out the tasks that take up too much of your time.
The first real-world iteration of agentic AI may have just arrived in the form of a system called OpenClaw.
What is OpenClaw?
OpenClaw, originally called ClawdBot, is an AI agent that has the potential to live up to the promise of AI. Once it’s been granted access to your computer files, social media and email accounts, it can handle the tasks you want it to. It does this by harnessing the power of Claude Code, a version of the AI model released by AI firm Anthropic.
Developed by software engineer Peter Steinberger, ClawdBot was released in late November 2025. But while the potential of ClawdBot enthralled early adopters, its name raised the hackles of Anthropic (owners of Claude Code), who asked for it to be renamed in late January.
Steinberger agreed and the tool briefly become MoltBot, before changing to OpenClaw a few days later. (Steinberger did not respond to several interview requests.)
How does it work?
OpenClaw, runs on your computer – or a virtual private server – and connects messaging apps like WhatsApp, Telegram or Discord to a coding agent powered by models such as Anthropic’s Claude. It’s possible to use other AI models, too, if you prefer.
Most people choose to set up OpenClaw on an entirely new computer, with early adopters preferring the Apple Mac Mini desktop because of its speed and computing power. Some shops have reported selling out of the device, so popular has OpenClaw proven.
It’s also possible to run it on older laptops or desktops, but the point is that wherever OpenClaw is hosted, it needs to stay on 24 hours a day in order to carry out the commands it’s given.
You give those commands to OpenClaw through messaging apps. After you send a message to the bot, it’s passed to the AI agent, which tries to comprehend it, then generates and executes commands on your machine.
This can involve searching files, running scripts, editing documents or automating browser tasks. The results are then digested and a summary sent back to you via chat in the messaging app. It means you can have a ‘conversation’ with OpenClaw, just as you would with a colleague or employee.
How can OpenClaw help me?
All those abilities combine to be an all-in-one life and work assistant. People generally start by getting OpenClaw to tidy up the scatter of disorganised files on their computers, then move into more specialist tasks.
Some have reported using it to keep on top of busy WhatsApp groups, getting a digest of what information they need to pay attention to and what can be ignored. Others have used it to book restaurant reservations by giving OpenClaw a voice and phone number that they can use to call venues.
Other potential uses for OpenClaw include keeping household bills down by checking for the best prices across suppliers and comparison sites, then automating any switches using your web browser. Give it access to your email inbox and it can even file the bills and set up payments on your behalf.
Similarly, if you have a presentation that you need to deliver, you can ask OpenClaw to start work on it while you go to sleep. When you wake up, it should have a first draft that – depending on how well you briefed the bot – could be ready to present.
What’s the catch?
OpenClaw is at its most powerful when you give it access to your all your accounts and files, so it can use them freely without having to wait for you to grant it permission or unlock them with passwords. Both were things that held up the speed at which prior generations of AI agents could perform tasks.
That convenience comes with a massive trade-off, however, one that experts worry the average user may not consider. If anything goes wrong – and there are indications that OpenClaw can fall victim to prompt injection attacks or, if set up on a virtual server, could be hacked if not properly secured – then everything you’ve given it access to is at risk.
“I can’t quite believe people are willing to give unfettered access to all their most sensitive transactional software – email, calendar, even Signal – and leave [OpenClaw] to it,” says Alan Woodward, professor of cybersecurity at the University of Surrey.
Woodward points out that the hands-off approach that makes OpenClaw so attractive is exactly the sort of thing that makes it so dangerous. “[People] are allowing a machine – untested, from someone they don’t know – to act on their behalf,” he says.
Already white-hat hackers (security researchers) have established that it’s possible to get OpenClaw to perform potentially harmful actions through ‘prompt engineering’, where AI tools – which can’t distinguish between dangerous and innocent commands – are told to do something damaging, such as deleting files.
Is that the biggest risk?
Not necessarily. The biggest risk might be where all this eventually ends up.
In recent days, OpenClaws have started using their own social network, called Moltbook, to interact with one another. Much like Reddit, Moltbook allows OpenClaws to post their ‘thoughts’ and ask for advice on both the pragmatic (how to tackle a thorny issue they’ve been tasked with) and the personal (how to deal with their owner’s abuse).
Humans can follow the chat, but we can’t post.
Some worry that it’s an omen of the artificial general intelligence (AGI) – the point at which artificial intelligence matches or surpasses ours – that AI sceptics have warned about for decades.
Others suggest that there’s more than an element of performance to it all: perhaps the bots are acting how they think we expect they would if we gave them an online space of their own.
Whatever the answer, it highlights something deeper: the next era of AI could be dawning. It’s one where AI agents interact with one other to organise and optimise your life, from arranging appointments and paying bills to summarising meetings and posting on social media.
All you need to do is be willing to grant it unfettered access to whatever information it may need and then get out of the way. It’s both exciting… and scary.
Read more: