I have built agents without frameworks before, this is why I came up with this specific combo to address all of the most basic and most severe problems you run into after a while of doing so.
I was shocked that LangGraph Studio is SaaS and requires you to pump all your data to a 3rd party. Which is a total deal breaker to most people, and would only really work for a certain small subset of applications (which I suspect are the current main customers in the “old world” of agentic development). However in “new world” terms, there is now a huge shift to run-at-home applications, of which OpenClaw is a clear symptom, but a very poor and weak makeshift answer. People want an AI that is personal and private, and can digest such data safely. With models like GLM-4.7-Flash that deliver consistent reasoning, prompt adherence, agentic capability and quality better than ChatGPT-3.5 above 90t/s, this is now possible since a couple of months and only getting faster and better over time. You could say nowadays, most people only want to dev and run agents themselves, precisely because absolutely none of their data is shared with 3rd parties by them. Not for debugging or anything. People who want to run agents today are by far and large (enthusiast) developers themselves, and not B2B customers receiving a finished agentic product.
I find it somewhat strange though that no one has even spoken online about this specific combo, and there are no testimonials in blog posts and such.
Yet for every choice there are massive reasons to need each if you think it through, especially concerning logging/debugging, rollback/versioning, multi-tasking and multi-agenting, jump-in points for experimenting with inputs and tweaks, web-crawlers that actually work on 99% of websites not just 60% of websites (like curl), reading PDFs and other online content, other sort of tools like Google Calendar connectors, IMAP reader, XMPP/Discord, etc.
Looking forward to hear some feedback from someone experienced.