With the constant stream of new LLMs, it’s easy to quickly forget recent history. This was proven once again with OpenClaw, the tool that allowed users to have AI models perform actions en masse. Today, this functionality is ubiquitous. OpenClaw 2.0, which came about “by accident,” is attempting to turn the tide with a “first-class browser experience.”
A browser experience, then, that’s how OpenClaw version 2.0 is being marketed, according to DevRel manager Hannes Rudolph. Every aspect of OpenClaw is being overhauled, from installation to messaging, memory, skills, models, and more. What was initially intended to be a cleanup of the project has become version 2.0 in just two months.
Ease of use
Version 2.0 is designed to enhance OpenClaw’s ease of use. This means users start with systems they’re familiar with, such as ChatGPT or Claude, and then go through a simple configuration process. They can then fine-tune their “Claw” via the chat interface. This is somewhat ironic, given that OpenClaw’s unique nature stemmed precisely from the fact that it allowed AI models to bypass this chat interface. The response to competing solutions is now to resemble these applications more closely.
OpenClaw’s goal remains the same. Workflows vary as widely as a computer’s applications. Think of reminders, automating repetitive tasks in the browser, or AI-driven email responses. The range of solutions doesn’t really capture the imagination, especially when compared to the competition.
The competition woke up
OpenClaw quickly became wildly popular shortly after its release in November 2025. Although even non-technical users were able to get started with the tool using some online help, the focus was explicitly on an audience that knew how to work with IT systems. The possibilities were impressive: it was AI “that actually does things,” in the words of its creator, Peter Steinberger. Chatbots were a thing of the past, and the promise of agentic AI, which had been a constant feature of presentation slides since 2024, was finally within reach.
This came with countless problems. Tens of thousands of OpenClaw instances turned out to be accessible via passive scans in January of this year; two months later, that number had risen to hundreds of thousands. Summer Yue, Director of Alignment at Meta Superintelligence Lab, noticed the negative impact of OpenClaw’s use when the tool deleted hundreds of emails from her inbox. Stories like this were commonplace early this year.
More importantly, OpenClaw’s strengths could be replicated. Steinberger himself soon moved to OpenAI. That company and Anthropic managed to give their AI models, Claude Skills and Custom GPTs, respectively, the same capabilities that had been achievable via the OpenClaw workaround. These options are more user-friendly, faster to implement, and cheaper than using OpenClaw with API access to other models.
Hype fades, adoption stalls
If we view GitHub stars and website traffic as indicators of adoption, OpenClaw has shown a leveling off since March 2026. However, exactly what that adoption entails remains to be seen. Hermes Agent, an alternative from Nous Research, appeared later but has garnered 239,000 GitHub stars, compared to 388,000 for OpenClaw. These figures at least show who was interested in such tools at a certain point in time, but they do not indicate how they relate to the use of agentic workflows via Claude, ChatGPT, or other business applications.
What is certain is that the hype surrounding OpenClaw has well and truly faded. That doesn’t necessarily mean it’s irrelevant. However, the feature set of version 2.0 is so conservative that we shouldn’t expect a resurgence from the competition to respond to it.
Read also: Why OpenAI hired the founder of OpenClaw