So I have been using OpenClaw for a week now

Hello 👽

Today we’ll be taking another spontaneous detour from our AI series for a Mid-Series Special 🤗. This time it’ll be about the highly popular OpenClaw project.

If you haven’t heard of OpenClaw, it’s an open-source tool/project that provides an agent that has several “skills” and can be taught new skills as well, which is increasing. Skills like managing calendars, browsing the internet, sending messages to social media platforms like WhatsApp, searching and summarising a user’s entire computer, etc.

Peter Steinberger, the creator of OpenClaw, was actually trying to build an AI-first summarisation tool and wanted to build an AI Agent that could help him with that task, but the Agent did so well, summarising his friend’s computer, that he figured there could be more to this, so he decided to teach the Agent more skills.

Before we go any further, I believe we all know what AI Agents are, right? If you don’t know or want to revise your knowledge, we already learned about AI Agents in our Mid-Series Special 3, AI Agents: The future of AI? This was over 2 years ago 😁, so you see, I am always at the bleeding edge of this AI technology 🤗. You just have to read that piece before you continue 😁.

Everyone and their sister in the tech industry would have heard about OpenClaw these past weeks, starting with the bots’ social network, where humans are not allowed, only bots. And the people are seemingly impressed by all the skills the open-source tool offers. I really was not one of those people, but as a scientist, I shouldn’t judge by bias, right? As they say, the taste of the pudding is in the eating 😊.

So I have been using OpenClaw for a week now.

Right out of the bag, OpenClaw had to mention the security implications of installing this tool, because it is truly immense. The installation process was pretty clean and straightforward. You get to choose the Large Language Model (LLM) model you want to use. I used Co-Pilot Pro here, which gives me access to several models right out of the box. The LLM you choose would definitely determine how good your agent is because the LLM model would be powering the agent. You also get to choose the skills you want the bot (Agent) to have, which include access to things like platforms like GitHub, WhatsApp, and so on. Most of the skills need you to login and give permissions. I also chose Terminal UI (TUI) for interacting with OpenClaw.

When you’re done with the setup, you get access to the Agent using text commands. You can also set up the interaction through Whatsapp, iMessage etc, so you can carry out the tasks right from your phone. Here are some JPEGS.

Even to browse, you need to setup Brave browser first, but it is able to browse direct URLs and get page data.

When asked to schedule a task in my calendar, it actually succeeded; I was able to see the task on my calendar. But it assumed the time in the first try (used 9 am), when I reminded it, it asked for the new time (I said 4:30 PST, which is 12:30 am WAT) and claimed to have updated it on my calendar, but I checked my calendar and the scheduled task was still for 9 am.

So here are my thoughts on OpenClaw

First, I think the hype is just, you know, too much. AI agents based on LLMs have existed for over 2 years now, and if you have ever needed an AI agent for a task, you can get it from several platforms, most of which are already open-sourced.

Then there’s the technicality as well. For now, only people with some decent technical background can be able to use it. Which brings me to the early point I made, that persons with a decent technical background who want to run some tasks with an AI agent can already find those open-source or paid agents. Because you have to keep in mind that most or all of the skills of OpenClaw are not new at all.

I sent a couple of WhatsApp messages to my friends from OpenClaw directly, and I had to provide their phone numbers manually, because it could not access names, even though I connected with the WhatsApp app (maybe because I was using the TUI instead of my WhatsApp app?). I then tried using Siri to send the same message with a voice command, and it was even more seamless, given that Siri is practically an obsolete AI assistant. I bet Google Assistant would 100x what Siri can do now lol.

Even on the OpenClaw website, reviews from people on how they are using OpenClaw are about scheduling tasks on the calendar and messaging with the supported social media platforms, ordering food, and the likes.

Finally, the security concerns here are real. AI agents would make mistakes (as you can see with my calendar scheduling task earlier), and given them access to your device and applications, they can shop for you, so they can access your financial data, etc.

But I understand the excitement, especially for people who are not familiar with the progress of AI agents so far. This is completely new to them.

Let me take us back to 2 years ago once more (2024), when the Rabbit R1 and Humane AI pin launched, I wrote a piece about them as well: AI the Product vs AI the Feature: A Case Study of Humane AI Pin and Rabbit R1 in our 4th Mid-Series chapter, told ya, I have been at the cutting edge of this technology 😎. These foremost companies already started using AI agents with their own custom hardware because of the strong belief they had on AI agents. The Rabbit R1 team even created what they called the Large Action Model (LAM) as opposed to LLM, which just generates text. LAM was supposed to be built with an Agentic workflow at its core, thus the “Action” term, because it wants to get things done, book flights, plan trips, order food, etc. None of which it was able to do well.

OpenClaw is doing what Rabbit R1 and Humane AI struggled to do, mostly because after two years, the technology has greatly improved, so it’s a matter of timing at this point. OpenClaw has been able to provide these capabilities to users right from their messaging platforms like WhatsApp, iMessage, etc., which is very cool.

Would AI agents like OpenClaw eliminate apps?

Peter, the creator of OpenClaw, said in an interview a few days back with Y Combinator that 80% of apps will disappear in favour of AI agents in the near future. You can watch the full interview here. This is the same thing, almost verbatim, as what Jesse Lyu, the founder of Rabbit R1, said and tried to achieve with his device. To eliminate what they see as the inefficiencies and friction of using apps.

You see, I really don’t agree with that idea. We are social animals. Interacting with some of these apps is part of the experience, as not all tasks are about efficiency. Not all interaction is friction. Sometimes the interaction is the value. Utility apps for scheduling tasks and messages, setting reminders and timers are arguably mundane and boring, so automating them seems cool, right, yeah. And I guess all smart mobile phones can do a good amount of this with their assistants with Siri, Google Assistant, etc. The percentage of utility apps is not much, and the time spent on them is even lesser. The popular apps like those for social media, games, music, fintech/bank, etc., which easily make up a majority of the most used apps, are not going anywhere, the experiential apps.

Would AI agents like OpenClaw eliminate personal assistants?

No, not really. I don’t think people who can afford personal assistants would want to be setting up their “automated” schedule themselves. There are AI companies with products that allow AI assistants to answer calls and pass it to a human if and when necessary. So let’s say the AI Agent can take your calls with your voice, schedule meetings using your calendar and so on, would that still eliminate Personal Assistants? Well, maybe Virtual Personal Assistants, right? But keep in mind these agents can make basic mistakes, but hey, so can humans too, right lol? 🤷‍♂️ But for now, I believe AI Agents like OpenClaw can complement Personal Assistants at best, the assistant to the assistant, as they say 😀

I guess I am more critical of the hype because I am at the forefront of the AI technology, as opposed to a lot of users who only know ChatGPT, Claude, Co-Pilot or what’s trending on the AI front. I have been following this field closely, I’ve seen IBM Watson, Boston Dynamics, Google DeepMind Projects and so on, all before Large Language Models.

OpenClaw shows promise of potential (I guess I can say the same for LLMs generally), but for me, that’s about it. We haven’t yet figured out applications for this technology clearly. In other news, the founder of OpenClaw, Peter Steinberger, has been hired by Sam Altman’s OpenAI, as reported a few hours ago 🙂. It’ll seem like not only Mark Zuckerberg knows how to poach talent in this fast-paced field.

See you in the next one 😎

⬅️ Previous Chapter