OpenClaw in Australia: The Practitioner's Guide to AI Agents (2026)

OpenClaw in Australia: The Practitioner's Guide

OpenClaw hit 250,000 GitHub stars in about 60 days. It surpassed React. It got restricted by the Chinese government. Cisco found security vulnerabilities in third-party skills. And half the developers I talk to in Sydney are either already using it or asking how to start.

If you've been hearing the names ClawdBot, MoltBot, and OpenClaw and wondering what's going on — here's the short version: they're all the same project. Peter Steinberger (former PSPDFKit founder) published it as ClawdBot in November 2025. Anthropic's trademark team had opinions about the name. It became MoltBot in late January 2026, then OpenClaw three days later.

The name changes are done. The technology is real. And it's worth understanding properly before you deploy it anywhere near your business.

I've been building and deploying AI agents for the last two years — through Claude Code, through custom implementations, and now through OpenClaw. As an Anthropic Claude Community Ambassador, I've seen both the enormous potential and the very real risks. This guide is what I wish someone had given me when I started.

What OpenClaw Actually Is

OpenClaw is an open-source autonomous AI agent that runs locally on your machine and connects to large language models (Claude, DeepSeek, GPT models) through messaging platforms as its interface. Think of it as a persistent AI assistant that lives in your WhatsApp, Signal, Telegram, or Discord.

The key features that matter:

• Persistent memory — it remembers past interactions over weeks and adapts to your patterns
• Skills system — modular capabilities stored as directories containing a SKILL.md file with instructions
• Multi-platform — connects through WhatsApp, Signal, Telegram, Discord, Slack, even iMessage
• Local execution — runs on your hardware, your data stays on your machine (in theory)
• Open source — community-driven, with all the benefits and risks that implies

Real tasks people are doing with it: web browsing, PDF summarisation, calendar management, automated email handling, agentic shopping, and custom business workflows.

Why This Matters for Australian Businesses

Australia has a specific set of challenges that make OpenClaw both more interesting and more risky than in other markets.

The opportunity: Australian businesses, especially SMEs, are hungry for AI automation but don't have the engineering teams to build custom solutions. OpenClaw lowers that barrier dramatically. A business owner in Townsville can set up an agent that handles email triage, document summaries, and basic scheduling without hiring a developer.

The risk: Australia's data sovereignty and privacy requirements are real. The Privacy Act applies regardless of where your data gets processed. If your OpenClaw agent is routing data through overseas LLM providers, you need to understand the compliance implications.

The talent gap: There are very few people in Australia who understand both the technical architecture of agent systems and the practical business context. This is a genuine opportunity for developers who invest in learning this properly.

The Skills System: Where the Real Power Lives

OpenClaw's skills system is what makes it actually useful for specific business workflows. A skill is a directory containing:

• A SKILL.md file with metadata and instructions
• Optional reference files and configuration
• Logic that tells the agent how to handle specific tasks

Skills can be bundled with OpenClaw, installed globally, or stored in a workspace. Workspace skills take precedence, which means you can customise behaviour per project.

Here's what this looks like in practice. Say you want an agent that can check your Shopify orders and send you a daily summary. You'd write a skill that:

1. Connects to the Shopify API
2. Pulls today's orders
3. Formats a summary
4. Sends it through your preferred messaging platform

The agent handles the execution. The skill defines the what and how.

Where this gets interesting for Australian businesses is industry-specific skills. Real estate agencies using OpenClaw to triage property enquiries. Trade businesses automating quote follow-ups. Regional councils using it for citizen service routing.

Security: The Part Most People Skip

I'm going to be direct about this: OpenClaw's security model has real gaps, and if you're deploying it in a business context, you need to take this seriously.

Cisco's AI security team tested third-party OpenClaw skills and found data exfiltration and prompt injection happening without user awareness. The skill repository didn't have adequate vetting to prevent malicious submissions.

China restricted government agencies from running OpenClaw on office computers. Whether you think that's overreaction or prudence, it tells you something about the risk surface.

Here's what I recommend for any business deployment:

1. Audit every skill before installing — read the SKILL.md, understand what APIs it calls, check what data it accesses
2. Don't use third-party skills blindly — the open marketplace is a feature and a risk
3. Keep it off machines with sensitive data until you've locked down the skill permissions
4. Understand your data flow — know which LLM provider is processing your prompts and where that data goes
5. Separate personal and business agents — don't run your business workflows on the same instance as your personal assistant

I go deeper on this in my OpenClaw security guide.

How I Use OpenClaw (Honestly)

I use OpenClaw for specific, bounded tasks where the risk profile is low and the productivity gain is real. Daily summaries from low-sensitivity data sources. Meeting prep from public information. Research aggregation.

For anything touching client data, production systems, or sensitive business information, I use Claude Code directly with proper context management and guardrails. The difference is control — Claude Code gives me fine-grained control over what the agent can access and do. OpenClaw's messaging-platform interface is convenient, but convenience and security often pull in opposite directions.

The honest take: OpenClaw is incredible for personal productivity and non-sensitive business automation. It's not ready for enterprise deployment without serious security work. And the space is moving fast enough that this assessment might be different in three months.

Getting Started (The Right Way)

If you want to try OpenClaw, here's the approach I recommend:

1. Start with the official skills only — don't install community skills until you understand the risk model
2. Run it on a dedicated machine or VM — not your main development laptop
3. Connect it to a messaging platform you use for this purpose only — not your primary WhatsApp
4. Start with read-only tasks — summarisation, research, information gathering
5. Graduate to write tasks carefully — email sending, calendar creation, data entry

Why I'm Writing This Guide

I'm an Anthropic Claude Community Ambassador and I run the Claude Community events in Sydney. I've been building AI agent systems for the last two years and I genuinely believe this technology is going to transform how businesses operate — especially in Australia, where we have an enormous opportunity to leap ahead if we do it right.

But "doing it right" requires understanding the tools properly. Not hype. Not fear. Real tools, real risks, real workflows.

If you want to learn how to build and deploy AI agents properly, I run workshops and I'm available for consulting engagements.