What this covers
This is a nine-minute explainer from Udacity in which Simon walks through what OpenClaw is, where it came from, and whether a typical viewer should bother with it. It starts from a definition: software that runs continuously on a device and bridges chat apps such as WhatsApp, Telegram, Slack, iMessage, and Discord to models like GPT or Claude. He frames it as plumbing rather than an AI in its own right, and notes that major vendors were already shipping scheduled tasks, file access, and locally running tools like Claude CoWork before it appeared. The video deliberately leaves out installation instructions.
The middle section covers the project's history and mechanics. It began as a personal project by Peter Steinberger, first called ClaudeBot, then briefly MaltBot, before becoming OpenClaw after Anthropic objected to the name. It is free, open source, and runs on macOS, Windows, Linux, and smaller devices like a Raspberry Pi. Right after setup it behaves like an ordinary chatbot, and it only becomes useful once it is connected to email, calendar, and files. It keeps memory locally and can fetch skills from ClawHub, which listed more than 15,000 at recording, or write a new one on request. Setup, he says, demands comfort with terminals and API tokens.
The contentious part is risk. Simon argues that an AI agent needs autonomous permissions to be effective, yet autonomy means "clean up my inbox" might become deleting everything. Downloadable skills add another attack surface for malicious code or data leaks. His mitigations include letting it draft but not send emails, blocking permanent deletion, restricting it to shared folders, and running it on a dedicated machine instead of a work computer. His verdict is that most people should probably not install it. The more durable skill, he says, is spotting tasks worth delegating and describing them clearly.
OpenClaw is an open-source AI agent framework that connects messaging apps to LLMs and external services, offering significant capabilities but presenting real security and autonomy risks that most users should understand before considering installation.
- OpenClaw is 'plumbing' that connects existing tools (chat apps, email, calendar, files) to AI models, not an AI itself
- Effectiveness requires autonomous permissions, but autonomy creates risk of unintended actions or data leaks
- The practical skill worth developing is learning to identify tasks suitable for AI delegation and describe them clearly
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
The core security challenge with AI agents like OpenClaw is that effectiveness requires them to have autonomous permissions to take actions, but if they can act autonomously, their actions may not be what the user expected, such as an instruction to 'clean up my inbox' being interpreted as simply deleting everything.
“Because you need to give it access to those external services and tools in order for it to be helpful. But by giving it access to those things, you expose yourself to risk. Because if I hook it up to my email, and then I tell my AI assistant, 'Hey, clean up my inbox.' Well, what does that mean? After all, a very simple way to clean up my inbox would be just delete everything.”
The ability to download new skills from the internet presents another security attack vector because malicious code could be inserted into new skills or code could lead to unexpected behavior or data leaks, which makes security professionals nervous.
“The fact that it can download new skills from the internet is another thing that makes security people very nervous because that's another vector of attack where perhaps something malicious may have been inserted into that new skill or just something in the code that could lead to unexpected behavior or a data leak.”
Risk mitigation strategies for AI agents include working with specific permissions such as allowing it to draft emails but not send them, preventing permanent deletion of data, or granting access only to files in specific shared folders.
“Sure, there are ways to manage some of the risk. You can work with very specific permissions, such as giving OpenClaw the ability to draft new emails, but not send them. Or to not have the ability to permanently delete anything. Or you might give OpenClaw access to only specific files that are placed in a specific shared folder.”
Even before OpenClaw existed, major AI vendors were already introducing similar functionality including chatbots that can run scheduled tasks, AI that can access files, and tools like Claude CoWork that run on users' own devices.
“Because even before OpenClaw existed, we were already seeing this kind of functionality start to appear from the major AI vendors. Chatbots that could run scheduled tasks for you, AI that could access your files, tools like Claude CoWork that can work directly on your own device.”
Initially after installation and basic setup, OpenClaw behaves similarly to a regular AI chatbot with access only to information provided during conversations, but its main utility comes from the next step of giving it access to external services like email, calendar, and files.
“At this point, after that initial install and setup, the basic behavior is going to seem very similar to using an AI chatbot. Because to begin with, all it has is access to the information you provide during conversations. It doesn't have access to your email or your calendar, your files, or any other data unless you explicitly let it do that or connect to those tools. But that typically is the next step you'd take is choosing what you want to give OpenClaw access to.”
Users should not install OpenClaw on their main computer or work machine, and should instead run it on an old or dedicated device; some users have been buying Mac minis as dedicated devices to run OpenClaw.
“Also, don't install it on your main machine or your work machine certainly. I'm running mine on an old laptop I wasn't using and famously a lot of people have been buying Mac minis as dedicated devices to run open claw on.”
OpenClaw is software installed on a device that runs continuously in the background and connects chat applications (WhatsApp, Telegram, Slack, iMessage, Discord) to AI models like GPT or Claude by acting as plumbing that bridges existing services and tools rather than being an AI itself.
“Openclaw is software you install first on your own device, your own computer... it runs continuously in the background... it lets you take the chat applications you already use, whether that's WhatsApp or Telegram or Slack or iMessage or Discord, and connect that chat application to an AI model like OpenAI's GPT or Anthropic Claude”
The practically valuable skill to develop now is the ability to recognize tasks in daily work that could be delegated to an AI assistant and clearly describe what you want the assistant to do.
“And the skill that is worth developing right now, whether or not you ever install OpenClaw, is the ability to recognize tasks in your day-to-day work that you could hand over to an AI assistant and clearly describe what you want it to do. That is the skill you can start practicing it right now.”
More important than installing OpenClaw is spending time thinking about what tasks you could delegate to an AI assistant, including browsing available skills on ClawHub, to identify use cases before worrying about technical implementation.
“What I think is actually more important than installing it is first just thinking about what you might do with it... So, spend some time thinking about how you might use it. Browse skills on ClawHub because someone who says, 'Okay, I know exactly what I'd like to use an AI assistant for. I'm just not comfortable installing and configuring it yet.' That person is in a better position than a developer who would say, 'Oh yeah, I could set it up in 10 minutes. I just can't think of anything I would ever ask an AI assistant to do.'”
The installation and setup process for OpenClaw is not user-friendly and requires comfort with terminal windows and API tokens, making it inaccessible to non-technical users.
“So, right now, the installation and setup is not user-friendly. You need to be comfortable working in the terminal window and dealing with things like API tokens.”
OpenClaw and similar AI agent tools will appear in users' lives and work very soon, making it valuable to understand what they are, what they are not, and why people are talking about them.
“OpenClaw is just the beginning. There will be more things like this and very soon.”
OpenClaw stores memory locally on the user's device and learns from conversations, continuously acquiring new abilities and becoming more useful and capable over time as it is used more.
“One of the key ideas with open claw is that it kind of starts to evolve. It stores memory locally on your device and it continues to learn from your conversations. It learns new abilities. It becomes more useful the more you use it.”
The video deliberately does not provide instructions on how to install OpenClaw.
“In this video, I am not going to tell you how to install Openclaw.”
For most people asking whether they should install OpenClaw, the answer is probably not to install it, though understanding what it is and how it works remains valuable knowledge.
“So, for anyone who is right now thinking, okay Simon, still sounds kind of interesting, should I install open claw? Well, if you're even asking me the question, my answer is probably not, but you can prove me wrong.”
There is a cliché of people who install and configure OpenClaw and then don't know what to do with it, suggesting that use case clarity is a common problem.
“Because there's almost this cliché of people saying, okay, I've installed and configured open claw, now what?”
OpenClaw was originally created as a personal project by developer Peter Steinberger, initially called ClaudeBot, then briefly renamed to MaltBot before being renamed to OpenClaw after Anthropic (makers of Claude) objected to the name similarity.
“This originally started as a personal project by a developer called Peter Steinberger, who then made the code available to download over GitHub. And he originally called it ClaudeBot. But Anthropic, who make Claude, were not happy with this increasingly popular project that sounded like Claude, so they asked him to change the name. And it was first renamed briefly to MaltBot, and then to Open Claw.”
OpenClaw can be instructed to find and install new skills from ClawHub, and if no existing skill meets a need, the user can ask it to create and code a new skill itself.
“You can just chat with your open claw assistant and tell it, go find this skill and install it. And if it can't find one that does exactly what you need, you could even ask it to create that skill and code it itself.”
Users should be very cautious about which external services they connect OpenClaw to and how much permission they grant it.
“And be very cautious about what you connect it to and how much permission you give it.”
In February 2026, headlines began appearing about OpenClaw as a new AI thing.
“In February 2026, a lot of headlines suddenly started appearing about this new AI thing called Openclaw.”
OpenClaw is an open-source project where people can view the code and contribute to it, is free to download and install, and can run on macOS, Windows, Linux, or other devices like Android phones or Raspberry Pi.
“It is an open-source project, so people can view the code, they can contribute to it. It is free to download and install. It can run on macOS or Windows or Linux. If you're nerdy, you can install it on anything, like an old Android phone or a Raspberry Pi.”
ClawHub is a website containing 15,000+ skills for OpenClaw (at the time of recording) that provide additional functionality, ranging from business-oriented skills like Trello and Salesforce integration to smart home skills like controlling Sonos speakers, with new skills being added continuously.
“There's a website called ClawHub that is full of skills for OpenClaw. At the time I'm recording this, it already has 15,000 skills. You can find everything from business-oriented skills like connecting to Trello boards or to connecting to Salesforce to smart home skills like controlling your Sonos speakers and there's more being added every day.”
There is significant public confusion and 'nonsense' being discussed about OpenClaw that the video aims to clarify.
“And there is a lot of nonsense being talked about Openclaw. Let's clear it up.”
