In case you haven’t heard (and you live under a rock) OpenClaw has taken the AI community by storm. As far as I can tell it provides the most autonomous AI assistant out there, from this open source repo. It’s different from using something like Claude Code because you don’t need to prompt it to do anything. Instead it runs constantly, learns about you through your email, Twitter account, Google Drive or whatever other tool you want to give it access to (there are 50+ native integrations) and does work on your behalf.

With Claude Code you are giving a directive, telling your assistant to build a new feature, or research a specific topic and wait for results. With OpenClaw you can simply tell it something like “improve my business” and it will give itself initiatives and give you deliverables as it finishes.

Let’s say you are an owner of an ecommerce store. You can set up OpenClaw and give it a vague instruction to increase profit margin. It will decide on its own what to do- analyze your conversion rate and best selling products, research competitors or Meta ads library and it may go in a direction you don’t expect- like write new scripts for UGC ads based on what it sees from a competitor store when you were thinking you just needed new landing pages. Then it might even start setting up an influencer outreach campaign. Basically, it will go in any direction you allow it access to.

That does introduce risks. You probably want to be careful with allowing it to have write/send permissions on your code repo, email or social accounts. Prompt injection can become real worry with a new source of cyber attacks being predatory prompt injection emails.

A Claude max subscription is $200/month. If you want to get the most out of running OpenClaw you probably want that so you can run Opus without maxing out usage. Then you have the deployment choice. Some people recommend buying a Mac Mini just for the purpose of running this, that would be $600 (or $50/month for 12 months). If you’re not as technical this is an option but way more expensive than a cloud deployment on Hetzner which you can do for $5/month. The repo has instructions for a straight forward process here.

So, theoretically, if you wanted to just try it you could get the lowest tier Anthropic sub for $20 and pay $5/month for the Hetzner server. That’s probably where I would start.

As with most AI tools the most important part for you to get right is the harness- the tools you give it access to. That is where most of the work for getting value from this thing is.

It’s recommended for it to have its own secret email account (so it can’t send accidentally on your behalf and is safer from prompt injections) and also have access to external coding tools instead of writing code itself to protect token usage. In this video here a user explains how he has his OpenClaw use Codex to write code instead of having it write code itself.

That’s it for now,

Alex