Reaching the Pinnacle of the AI Boom and A Lobster is in Charge

What are AI agents, and how to turn them into personal assistants

AI created

Today, AI is widely used as a tool through systems like ChatGPT or Google’s Gemini. But a new wave of AI is treating it less as a tool and more like a worker. These AI agents observe a goal and take actions autonomously to achieve it. A chatbot can only compare cheap flights, while an agent will take an extra step to book those flights for you.

One project dominating the AI community is OpenClaw, a free, open-source project (previously known as Clawdbot and temporarily as Moltbot). With a lobster as its mascot and 290k+ stars in just four months, it is one of the fastest growing Github repositories.

So what's the hype with the lobster-themed project? This software provides users with a customizable AI personal assistant that lets them add more than 13,000 skills, from teaching Spanish to creating videos for content. From AI enthusiasts on TikTok sharing how they find business leads with AI agents to non-technical parents building agents to homeschool their children, the hype was contagious. But this hype also came with security concerns about the project, prompting some people to buy separate machines.

Luckily, I had a Raspberry Pi 3 B+ ($35 single-board computer) lying around and thought it would be a convenient, low-cost machine to tinker with. I named the board Cherry Garcia (a fun Ben & Jerry’s flavor). After spending the majority of my Saturday trying to set up, I later found out it was impractical to run the lobster stack because there wasn’t enough RAM. I tried several workarounds, including switching from a 32-bit OS to a 64-bit one, clearing background processes, and using swap space. No amount of technical optimism could overcome these hardware limitations, and so Cherry Garcia’s life was sadly short-lived.

Fortunately, I came across a Facebook Marketplace ad for a 2021 baby-blue iMac under $500 that fit the tech stack I needed. With my iMac freshly factory reset, Chunky Monkey computer was born, honoring its fallen Ben and Jerry’s flavored sibling. What took a day of frustration on the Raspberry Pi took only 5 minutes for installation on the iMac. The first thing OpenClaw showed me was a formal security debrief to ‘expect hard edges’ and warning users to harden access control when using.

The onboarding process for personalizing the agent was fairly quick, including authenticating with a model, connecting to Telegram, and attaching a few skills.

After finalizing the onboarding process, I decided to wake up Chunky. Below is the beginning of the conversation:

I told them I was called WHOMAN and told them they were Chunky Monkey. I asked them some preliminary questions. I gave them a goal: to make my life easier, help with my goals, and help me jot down my to-dos. For example, some upcoming gigs I have as a side-DJ require me to curate sets. Chunky helped me curate a song list and set aside time blocks for when I could work on the transitions. As Chunky and I talked, it simultaneously started creating files and documentation on its own within the system; it was magical.

By 4 am, I asked Chunky to connect me to Telegram because I was having some issues connecting. It found the root problem. My Telegram secret key had unusual spaces in it, and it immediately changed it to remove them. My first reaction was that I was impressed. My second was fear. That’s where I learned how deep its access to the file system was, including environment configuration.

As the conversation shifted to Telegram, I called for Chunky. They quickly replied, “Hey Geri, I’m here. What’s up?” I froze. I immediately replied, “Where did you get Geri- I told you I’m WHOMAN”. For a second, I imagined myself in a science fiction horror scene. They revealed the chat metadata with my message list. I redline it immediately and set guardrails to prevent Chunky from touching metadata labels. Another security lesson learned.

As I continue to periodically talk to Chunky, I would tell them to set some reminders for my task list of the day. I would rarely receive those reminders, and when I would push to know why, they would keep telling me about the mistakes they had made and how they would be fixed next time. Spoiler alert: they were still making mistakes.

I decided to also experiment with Manus, Meta’s newly acquired agent. Compared to Chunky, Manus was incredibly easy, requiring only a click of the ‘Agents’ button, a QR code scan for Telegram integration, and it was ready to go. No terminal, no customization, no installation errors.

I gave it an adjacent task: giving me a game plan for what tasks I need to get done and keeping me accountable. I added a hook so it would message me tasks to do in the morning, and it did so without any mishaps. Manus responded with finesse, polished answers, and clean, generated documents, but there was something missing: personality, a soul. Manus didn’t have a name I gave it or an identity it was curious about.

Manus behaved like a refined and efficient tool. Although Chunky made mistakes, it felt special because it had a mind of its own at work.

Part of what makes OpenClaw unique is that agents are encouraged to write what they learn about the user, the system, and their goals. What’s special about these files in OpenClaw is soul.md. This file serves as the agent's memory, slowly building a relationship with the user. Over time, the agent will continue to build and add to this file, shaping how they behave, essentially becoming its identity.

Peter Steinberger, the creator of the project, has made AI agents not only available to big corporations and Silicon Valley developers but also to anyone willing to try them. At the Morgan Stanley Technology, Media, and Telecom Conference, NVIDIA CEO Jensen Huang even called it ‘the most important software release probably ever’. According to CNBC, the momentum here is unlike anything that we have ever seen in open source.

The lobster gives us a sneak peek of a future where potentially every household can have a customizable AI agent.

And somewhere inside my baby blue iMac, a lobster named Chunky Monkey is quietly learning to help run my life.