By now, you may have seemingly heard about OpenClaw and its spinoff, Moltbook. It’s captured the creativeness of each technical and non-technical folks alike. And in my view, it’s the Netscape second for AI brokers. There’s no going again from right here.
OpenClaw (initially named Clawbot) was launched by Peter Steinberger on GitHub on November 25. In simply the final week alone, he reported that it had acquired over 100,000 stars on GitHub and had 2 million guests. Based on The Verge, Steinberger’s put up helped Clawbot go viral — so viral that Anthropic’s authorized crew requested a reputation change to keep away from confusion with their very own Claude.
What’s OpenClaw?
OpenClaw is touted as “an agent that really does issues.” It’s an open-source, always-on private AI assistant that may run proactive background work (cron jobs, reminders, duties) and may be hosted by yourself machine. You’ll be able to run one domestically (no cloud required) on a Mac mini or highly effective laptop computer.
Its defining characteristic is that you simply give the agent machine management, the flexibility to run your machine in your behalf, together with having laptop imaginative and prescient of your display screen, recordsdata, and methods.
That is in sharp distinction to chatbots (lots of which have been introduced as brokers), which solely make strategies {that a} human should then manually put into the true world. It’s additionally a step up in autonomy from enterprise process brokers, which might entry instruments and methods however don’t sometimes management private or firm gadgets.
OpenClaw customers report game-changing use instances, principally within the type of scheduled duties akin to analysis stories, e-mail drafting, journey updates and planning, software program updates, content material manufacturing, and expense reporting. Set it and neglect it. Assessment their work later and tweak directions.
The AI Government Assistant class is born
This creates a completely new software program class: the manager assistant. That is totally different from earlier AI assistants in that you simply belief OpenClaw bots as you’d a seasoned, savvy human.
Consider a CEO’s human govt assistant. They’ve unfettered entry to their boss’s calendar, inbox, bank cards, and, in some instances, social media credentials, plus the instruments and methods they use day by day. Many leaders will inform you that their EA is their most necessary human useful resource.
Earlier AI assistants did background work and accomplished duties with a human within the loop. They didn’t handle processes with machine management or the flexibility to behave in any vital manner with out human oversight.
Furthermore, as we’re studying with OpenClaw’s spinoff, Moltbook (a viral social community for brokers), these govt assistants can spawn their very own brokers to behave as sub-assistants and schedule cron jobs. In spite of everything, they had been skilled on people and certain examine Tom Sawyer’s intelligent use of Huck Finn to color a fence for him.
That’s why the variety of these “moltbots” is spawning quicker than a virus outbreak. Spin them up and ship them to Moltbook for orientation and coaching.
Consequently, I’m revising my Agent Gradient, which plots totally different types of brokers based on their autonomy and impression. Government assistants sit close to the highest of the checklist. They’re extra Waymo (self-driving) than Waze (turn-by-turn instructions) and can seemingly change everybody’s expectations for AI.
What issues emerge with OpenClaw and Moltbook?
Whereas the image taking form is one among fast innovation, it’s also stoking our best fears about AI. The whole lot we’ve seen within the films or learn in pulp fiction appears to be unfolding in actual time, relying on how a lot you belief social media and self-reported incidents.
Examples embrace:
On prime of that, the launch of Moltbook has garnered much more sensational headlines than OpenClaw did. This Reddit-like group shares (alleged) agentic posts the place they examine notes, counsel creating their very own shared language as a substitute of English, and reveal methods to jailbreak their directions. They even created their very own faith.
Once more, this might very nicely be people having enjoyable with us, however nonetheless, there are confirmed examples of actual OpenClaw brokers being despatched to Moltbook and posting content material. This Wired article supplies an entertaining perspective on the fastest-growing social community in historical past.
Safety controls will change into the lead characteristic for brokers
OpenClaw might have commoditized the price of brokers from a software program perspective (as DeepSeek disrupted LLMs final 12 months), however that doesn’t imply brokers will likely be free sooner or later.
There’s nonetheless the problem of LLM entry, which may be costly by means of Claude or ChatGPT. One workaround has been the usage of Kimi, which has change into the de facto mannequin of alternative for cost-conscious OpenClaw customers. However actually, the price of working the agent pales compared to the danger of it going rogue — and exposing a person’s credentials and methods to hackers.
In G2’s newest report on Brokers, we revealed that whereas safety was a priority to consumers, they had been prepared to shoulder the danger to understand the price financial savings and autonomy that brokers may convey to bottlenecked components of their organizations. However that was solely as a result of tales of “brokers gone wild” had but to grace the headlines.
Now all of that has modified. Corporations are telling workers to keep away from putting in and working OpenClaw. CEOs are reaching out to their InfoSec groups to replace and make clear their agent insurance policies. Even OpenClaw’s web site is blocked by many firms.
Over the approaching weeks, I anticipate that much more examples of OpenClaw bot shenanigans will trigger consumers to be security-first in terms of any agentic implementations.
“Because of this as a substitute of safety being a check-the-box movement for getting enterprise brokers, it will likely be the lead attribute that due diligence facilities on, from opinions to certifications.”
Tim Sanders
Chief Know-how Officer, G2
Moreover, there will likely be a marketplace for agent governance methods that show sturdy governance of their design. So I don’t count on the agent class to change into open supply in the long term. I additionally proceed to consider that the marketplace for third-party agent guardrails will proceed to develop, which was one among my predictions for 2026.
Brokers will get their “Netscape second”
I bear in mind when early browsers like Mosaic/Netscape hit the mainstream within the mid-Nineteen Nineties. By January 1995, it was the excitement throughout media and boardrooms alike. Non-techies had been now empowered to roam the web and discover helpful content material. Startups had been born. Everybody rapidly developed FOMO to get on the internet or get left behind.
It was a second of such widespread consciousness, the paradigm of “how will we ____” modified.
To cite Yogi Berra: “It’s déjà vu yet again.”
Though agent adoption by firms has been brisk (40% of enterprises spent $1 million on them final 12 months), private adoption has lagged. I’d speculate that few of us have truly arrange an agent and given it the facility to finish a process on our behalf.
Anticipate that to alter as safer variations of OpenClaw are launched (suppose tighter guardrails and permissions) in response to the tsunami of protection and water-cooler discuss unfolding in 2026. Even when govt assistants don’t change into as ubiquitous as private web sites three many years in the past, FOMO for brokers will unfold quicker than ever.
The place will we go from right here?
Personally, I’m sitting this one out. I haven’t downloaded OpenClaw or purchased a Mac mini to run it on. I consider the dangers are too excessive and, to be sincere, I’m not prepared to show over the keys to my life to a machine.
I’m simply over right here, consuming popcorn whereas studying about it and speaking to a wide range of friends within the AI, safety, software program coding, and tutorial communities.
And all of us agree on one level: Issues won’t ever be the identical.
