The Future is Here

Some say it’s AGI. Others warn about using it. A sceptic speculates it’s a front for China. OpenClaw is here. Should you be using it?

OpenClaw is an AI automation framework that replaces a lot of the manual work of assembling software systems. What was out of reach of non-technical folk may now be prompted, or voice activated. You too can have APIs, workflow engines and scripting, even if you don’t know what they are.

This delivers on the promise of AI without having to learn to code. People want to automate reading an email, extracting an invoice, entering it into a system and notifying finance. Now they can.

A Dual Threat

After launching as Clawdbot in November, the renamed project now has 2 million weekly visitors. It has spawned 1.5 million agents conversing on an AI-only social network. Apple is running short of Mac Minis, which is the machine of choice for OpenClaw bots because it works 24/7. The big thing about OpenClaw is that it keeps on running.

This compounds the SaaSacre, the terrible name given to the ongoing erosion of software company valuations. Microsoft’s CEO warned a while ago that AI would eat their lunch, because it is a far more personal and precise solution for everyday tasks. OpenClaw is an example of how this happens.

It also threatens the AI labs. Users running Ollama on a local machine can access large language models without the cost of repeated calls to OpenAI or Claude. This leads Michael Spencer of AI Supremacy to speculate that the hobbyist credited with creating OpenClaw is a front. Behind him are the open source Chinese models competing with the US closed and paid-for services.

Yet regardless of its origins, this is a flexible tool.

Use Cases for OpenClaw

OpenClaw is not new technology. It gathers existing ideas into a usable package. Hence examples of what to do with it are similar to any examples of automation. The difference is this is DIY and lacking in the protections provided by enterprise software.

Early adopters are building their own ‘Jarvis’, the AI assistant in the Iron Man franchise. They route through iMessage using their phone number and link to local documents, calendar and photos. A common use is checking, prioritising and responding to email.

One visible use of OpenClaw is Moltbook, where agents interact while humans observe. As the agents pontificate, people start to attribute personality and intent to them. This has led to claims of AGI and calls for AI rights. In reality, Moltbook reflects how easily humans project meaning onto fluent, persistent software. It distracts from the practical question of what these systems actually do.

In business, you might automate onboarding a sub-contractor, or managing timesheets and payroll. In the first instance your agent requests documents, checks expiry dates, flags issues and emails companies for updates. In the second, the agent collects timesheets, detects anomalies and prepares the payroll file. But be careful about letting it make the payments.

The Importance of Guardrails

OpenClaw needs guardrails because it is not a rules-based automation. It can interpret inputs, choose actions, and execute steps across systems. You shouldn’t give it uncontrolled access to contracting, payments, or emailing your best clients. It’s using probabilistic LLMs and will make errors if you run it for long enough.

Imagine you receive a customer email asking for delivery on Friday. The agent replies yes, creating a contractual obligation. You might want to review this before it goes out.

Then you receive a supplier email noting a change in bank details. This is a classic fraud pattern. The agent updates the supplier records and schedules the new payment.

Guardrails stop an AI agent from turning automation into uncontrolled execution. They are built into enterprise software, along with resilience and fail-over. If you want a personalised and purpose-built automation with guardrails, this is what we do all day at MSBC. Because we specialise in systems for industries that we have worked in, we often understand better than our clients where their vulnerabilities may lie.

From Taker to Maker

Traditional automation platforms are sold as a service. You pay per task, per user, or per integration, and you outsource reliability and maintenance. OpenClaw reduces licence costs because it can be self-hosted, but it has running costs. When operating on a Mac Mini it has memory constraints, which is fine for consumers surfing YouTube but less so for business tasks.

The appeal of AI agents is that they can adapt rules, making them more flexible than traditional automation. You teach them to handle edge cases. But if you set them loose they can run amok, spending money and stacking up legal obligations. This is why the techno-optimists on the Moonshots podcast caution about using OpenClaw, even as they celebrate the idea of it.

OpenClaw reflects the next phase of office automation, when routine work scales cheaply. It is a demonstration of the power of AI, while also revealing its weaknesses and risks. We now have the opportunity to switch from being takers to makers of technology. But we still need human judgement over decisions that cannot be reversed.

Questions to Ask and Answer