Not sure if this will surprise you - but I 100% agree with this. I went through the journey that many others did - implementing the loop, then trying to make it useful, realizing the limitations, etc.
I came to similar conclusions - what does valuable agentic software look like? It's not OpenClaw (yet)
The game theory then, in my opinion, is to focus on the knowable frontier - implement tools we can trust - and continue working and sharing that work.
I am holding onto the optimistic case - valuable use cases beyond coding agents will emerge.
I don't have a good answer - I've seen a lot of agent deployments but the space is evolving quickly and it's difficult to meaningfully discuss patterns.
This will be solved - and I hope that Jido can be a meaningful participant in that wider conversation.
My strongest opinion with Jido is that agents must be architecturally correct WITHOUT LLM's before they can be correct WITH LLM's
Jido core has zero LLM support for this reason.
There's nearing 40+ years of "Agent" research in CompSci, LLM's came along and we threw out all of it. I didn't like that so I spent time researching this history to do my best at considering it with Jido.
That said, I love LLM's - but they belong in the Jido AI package.
The core of Jido will run on a Raspberry Pi - we've even had people look at running Agents inside the BEAM where the BEAM is deployed on bare metal (embedded)
https://jido.run/docs/getting-started/new-to-elixir