Back to blog
3 min read
OpenClaw: The Most Revolutionary and Useless Repo on GitHub

OpenClaw is absolutely useless and revolutionary at the same time. I hadn’t tried OpenClaw until early March because it’s just an agent like we’ve been building for years now. Sure, they get smarter and smarter, but the core remains the same. Yet somehow, OpenClaw managed to literally explode with over 339k stars today. Judging by the stars, it’s even more popular than the godfather of all repos, React.

But let’s ignore the stars, they are an interesting anecdote for my next point, but they don’t matter. OpenClaw showed what agentic systems could be capable of, even for people who can’t code. Sure, for me it was cool as well, but I could have built the pipeline I needed with some vibe coding and Langchain in an evening. What made it so powerful is the clever usage of Memories (aka markdown files) and skills. Some skills and features are prebuilt so the agent can be talked to via Slack, and just by prompting, it can get access to its own mail account and legitimately start to act like an employee: receiving emails, summarizing today’s learnings, categorizing to-dos, forwarding emails to the correct person, querying the knowledge base, and drafting responses. These micro-interactions are so incredibly powerful with agentic systems.

Problems arise as soon as you try to scale further and further. The context gets bloated really fast, it keeps forgetting stuff, and skills are nice but are non-deterministic by nature. For many use cases, you know that you want that skill being invoked every time, the agent might now use it though. It lacks deterministic capabilities. In the end, it comes down to pipelines. When we figure out how to embed non-deterministic agentic AI capabilities within deterministic pipelines, these things will blow up. As of March 2026, it’s still a very technical job to keep the context clean, receive the correct memories, and invoke the correct actions.

That’s the problem with OpenClaw: it requires a lot of context management, memory management, and different skills for different use cases so that it becomes a jack-of-all-trades and therefore gets worse and worse. It’s great, but OpenClaw has to evolve. Even if it does not, it showed us that the world will jump on any advancement to the OpenClaw foundation as soon as possible.