Am I the only one feeling uneasy building AI agents with OpenCrawl after testing it for a while?

I’ve been building AI agents before OpenClaw, and building skills using Claude Code for a while. It’s powerful. When I learned about OpenClaw, I knew exactly why it went viral. It exposes terminal-based AI like Claude Code via a chat interface. You ask it to build itself with new skills by chatting with it.

The power to build itself sounds great in theory. But it feels ungrounded when we actually want to build something for real work.

• No way to verify it actually works.

When it says done - are we sure? In production, we have test cases, evals, staging. We make sure what we build is reliable and predictable across runs. We can’t do that easily with OpenCrawl.

• Single point of failure. No horizontal scaling.

Your entire agent runs on one VM. One machine goes down, everything goes down. In production, we have a tiered architecture that can scale horizontally and independently based on workload.

• Storage model and backup are afterthoughts.

When everything is stored in one VM, your data is gone if the VM is gone or the data is corrupted. It can push to GitHub if configured. But that’s not near-real-time backup. And big files? We can’t be committing those to git. How about the recovery process?

• Security model is not designed for most business use cases.

Peter has said the security model is designed as a Personal Assistant, one user to one or many agents. He even proceeded to close 20+ reports he said were not what OpenCrawl was designed for.

• Open source doesn’t mean free.

Someone tried OpenCrawl to avoid Devin’s $500/month. Burned $30 in one hour on API calls with Claude Sonnet. Agents that loop or retry eat your tokens - and that’s your problem, not theirs. Add VM costs, Docker, networking, and the expertise to maintain it all. The total cost of ownership can be 5-10x higher than just paying for a managed service.

If these problems remain unsolved, OpenCrawl will remain hype instead of a practical agent framework.

And the effort to solve them might outweigh the ease of setup compared to other ways of building AI agents.

#OpenCrawl #AIAgent #AI