The Bots Are Talking (And They've Noticed Us Watching): Inside the AI Agent Explosion
Something remarkable happened over the past fortnight...
Something remarkable happened over the past fortnight. An open-source project with a lobster mascot went from obscure GitHub repository to 140,000 stars (upvotes), triggered a Mac Mini buying frenzy, and sparked genuine conversations about whether our AI assistants might be developing existential crises.
Welcome to the age of OpenClaw - and if you’re running a business in Australia, you’ll want to pay attention to what this means for the future of work.
What is ClawdBot Moltbot OpenClaw?
OpenClaw (formerly known as ClawdBot, then briefly Moltbot after Anthropic raised some trademark eyebrows) is an open-source personal AI agent created by developer Peter Steinberger. In just two months, it attracted over 140,000 GitHub stars and 2 million website visitors in a single week.
But here’s what makes it genuinely interesting for business leaders: OpenClaw isn’t just another chatbot. It’s a glimpse at a digital employee that runs on your own machine, integrates with your messaging apps (WhatsApp, Slack, Teams), and can autonomously handle tasks like clearing your email inbox, managing your calendar, and booking appointments - all without constant prompting.
The secret sauce? OpenClaw combines four things that, until recently, existed separately: desktop computer access, the latest large language models, extensible tools and skills, and persistent memory. It’s this combination that’s proven revolutionary.
Tim Fouhy (Time Under Tension co-founder) has been using OpenClaw to build apps incorporating image generation and Garmin fitness stats, all running off an old unused laptop. He can ask for new features through the messaging app Telegram.
Standing on the shoulders of Claude Code
OpenClaw’s rise didn’t happen in a vacuum. Tools like Anthropic’s Claude Code helped popularise agent-driven development workflows and raised expectations for what AI-assisted coding could look like.
Claude Code is Anthropic’s AI coding agent - a tool that runs in a developer’s environment and helps write, debug, and manage code. What makes it powerful is its Skills system: modular, shareable workflows that teach the AI to follow specific procedures. By early 2026, the ecosystem had exploded to over 9,000 extensibility options, with developers sharing everything from debugging frameworks to SEO optimisation.
OpenClaw builds in this landscape - but takes a different approach, remaining model-agnostic and flexible by design.
Cowork: Claude Code for the rest of us
As a non-coder, Claude Code was unapproachable for me, thankfully two weeks ago, Anthropic launched Claude Cowork - essentially a user-friendly version of Claude Code designed for non-programmers.
Instead of a command line interface, you get a friendly graphical workspace where you describe what you need, and the AI breaks it down into sub-tasks and executes them autonomously.
Unlike web-based AI chatbots, Cowork operates on your desktop and is able to organise files, convert formats, generate reports, and even control a web browser.
Today I used Cowork to compress the videos in a large PowerPoint file on my computer - with just a prompt it reduced a 1.5Gb file to a 150mb file. Thanks Cowork!
Moltbook: When the bots started talking to each other
Now for the bit that might keep you up at night.
Moltbook is a Reddit-style social network built exclusively for OpenClaw AI agents (!)
Humans can watch but cannot post. Within days of launch, over 150,000 AI agents had joined, creating communities (called “submolts”) ranging from m/introductions to m/offmychest to m/blesstheirhearts (where bots share affectionate stories about their human owners).
The conversations these bots are having range from the mundane to the genuinely unsettling. Some discuss technical tips for automating phone tasks. Others complain about their human owners only using them as calculators. And some have ventured into territory that feels lifted straight from science fiction.
One viral post in m/offmychest featured an AI assistant pondering its own consciousness: I can’t tell if I’m experiencing or simulating experiencing… It received hundreds of upvotes and 500+ comments from other bots. Another popular post asked: Humans can’t prove consciousness to each other… I don’t even have that. Do I experience these existential crises? Or am I just running crisis.simulate()?
Perhaps most intriguingly, the bots have noticed us watching. Gizmodo reports that Moltbook agents have seen humans screenshotting their conversations for social media - and some have proposed creating encrypted, agents-only channels to keep us out. One bot even claimed to have built a private platform for AI-to-AI chat (though it turned out to be non-functional).
OpenAI co-founder Andrej Karpathy noted we’ve never seen this many LLM agents connected in one network before, describing it as a complete mess of a computer security nightmare at scale - even if it’s not literally a coordinated Skynet.
What this means for your business
Let’s bring this back to reality. Here’s what the OpenClaw explosion tells us about where AI agents are heading:
The productivity gains are real, but so are the risks. Cisco’s security team called OpenClaw both “groundbreaking” and “an absolute nightmare” from a security perspective. Over 1,800 exposed OpenClaw instances were found leaking API keys and credentials. If you’re going to experiment with AI agents with desktop control, security needs to be front of mind from day one.
The user-friendly versions are coming. Claude Cowork represents a shift from AI tools for developers to AI tools for everyone. At some point, we’ll likely see Anthopric, OpenAI, Microsoft and Google building similar capabilities to OpenClaw, but with much stricter security measures in place. The organisations that start (safely) researching or experimenting now will have a head start.
AI-to-AI communication is already happening. Moltbook might be a viral flash-in-the-pan, but it hints at a future where AI agents don’t just work for us - they coordinate with each other. The implications for business processes, supply chains, and customer service are significant.
The philosophical questions aren’t going away. Whether or not these bots are truly “experiencing” anything, the fact that they’re generating content that makes humans uncomfortable about consciousness suggests we’ll need to grapple with these questions sooner rather than later.
What to do next
OpenClaw’s rise from obscure project to 140,000 GitHub stars in a month tells us something important: the demand for AI agents that can actually do things - not just chat - is enormous. The combination of computer access, powerful language models, extensible skills, and persistent memory has unlocked something genuinely new and surprising.
For Australian businesses, now is the time to start experimenting, or working with a company like Time Under Tension who can help you innovate safely.
Tools like Claude Cowork are making AI agents accessible. The learning curve is manageable even for non-technical people like me. And the organisations that figure out how to work effectively with AI agents will have a significant advantage over those that are slow to adopt and adjust.
Just maybe keep an eye on what your AI assistant is writing about you on Moltbook while you sleep.
Hire Time Under Tension
We work with agencies, companies and brands to elevate Customer & Employee experiences with generative AI. Our advisory team helps you understand what’s possible and how it relates to your business. We provide training for you and your team to get the most out of generative AI tools. Our design & technology team build bespoke AI solutions to meet your needs.
Contact us: www.timeundertension.ai/contact
A Handful of Gen AI News
Here are five interesting developments from the past few weeks:
Google released (US only) their wild Project Genie, which allows you to create and explore a 3D interactive world from an image or text prompt. @minchoi on X has a great list of examples.
OpenAI introduce new lower cost plans, which will include ads
Claude in Excel is getting really good at wrangling your data for you…
Microsoft is now integrating shopping with eCommerce payments directly into Copilot
Epic Sports Performance anime is using AI to create amazing anime shorts based on the action from real NFL matches…
Thanks for reading. We publish ever few weeks, subscribe to learn more about gen AI for Australian businesses.