Artificial Intelligence – Closed Claw

OpenClaw will be a victim of its own success.  

• The latest fad sweeping the AI industry is Open Claw, which allows agents to connect to all of one’s apps and services, but its open nature ensures that no sensible person would ever use it, opening the way for a series of closed and proprietary versions to take over the market.
• OpenClaw is an open source platform that works as a gateway to allow AI models and assistants to connect to the apps, data and services that users access every day on their personal computers.
• In effect, it allows AI assistants to do things on one’s behalf very much as a human executive assistant would.
• The project has gained a lot of traction with an estimated 2m users, and the community has built 13,700 extensions, which confer other abilities such as stock market trading and smart home controllers.
• This sounds great, but to work well, it needs high-level access to one’s personal data and accounts, which makes the system very insecure.
• This is because LLM-based services are inherently unreliable due to their inability to understand causality, and there is nothing to stop the agent from emailing your personal data to scammers or booking travel that makes no sense, which then costs a lot to reverse.
• Consequently, no rational human or enterprise would do something as dangerous as this anywhere near their personal data, meaning that, as it is, it is not of much practical use.
• However, if it can be managed and controlled reliably, then the security issue goes away and something very useful remains.
• What I suspect we are about to see is a series of product launches that are aimed at taking what OpenClaw has pioneered and productising it in a way such that it becomes both safe and useful.
• This is why OpenAI has hired the creator of the OpenClaw tool, and I suspect that we will soon see OpenAI launch a product that has a series of built-in controls for consumers.
• I also expect Anthropic to quickly extend its desktop tools in this direction.
• There is also chatter in the market that Nvidia will launch something similar within its Nvidia Inference Microservices (NIMs) products that allows enterprises to connect all of the NIMs they have created to their corporate systems.
• The other ones to watch very closely are ServiceNow, UiPath, Salesforce and so on, which are likely to create offerings that match the use case but also give enterprises complete control and security over their data and systems.
• This is the battle for the AI ecosystem that RFM has written about on several occasions (see here), where the goal will be to become the platform that enables users to access AI easily and safely.
• Here for OpenAI, the consumer is a must-win, if the company is to survive, and competition is likely to be fierce with Google, Meta and probably Apple at some point, all vying to do the same thing.
• In the enterprise, Anthropic is certainly a contender, but I think it is quite possible that those that already have hooks into the enterprise, such as ServiceNow, Salesforce, IBM, SAP and so on, are in the best position.
• This is also because they can sit on top of anyone’s models, giving their clients access to all of the AI that they want from a vendor they already know and trust.
• Anthropic’s ongoing fight with the US Department of War and its designation as a supply chain risk, whether merited or not, will further encourage enterprise users to keep it at arm’s length.
• This is just another reason that is likely to push enterprises towards their existing vendors who are embracing AI as opposed to going direct.
• Hence, I think that the AI-related sell-off of much of the software and services sector is greatly overdone in many, but not all, cases.
• Here, my top pick is ServiceNow, where I already have a position and where I am looking for $233, which is 89% above current levels and 117% above where I took my position.