I always found Erlang (and Elixir) to be an interesting use case for this kind of agentic harness. It can run distributed across machines and can hot swap any function. It's order of magnitudes faster to update, and often easier to introspect and keep compatible over time.
I built something for this and also found that the idea of a "guardian" core that can detect misbehavior (crashing etc) and roll back a messed up state really helped. But it's a lot of work to get a harness to behave well, and each model reacts differently to the harness limitations, so I found more success in building a meta-harness instead (just run existing harnesses inside a common UI and inject all capabilities).
Ok i should not be trying to wrap my head around this, this late at night - but i got to be frank and tell you this is way too loud and indecipherable. I need a two sentence explanation given that i work with `pi` and build my own tools.
Do i have it right that its some prompts injection for reflection and a composition scheme for the prompt/context where each element is version controlled?
If so, my experience with all the SOTA models is that they're pretty bad at finding the 'clean' solution to self improvements; eg they still always add text instead of taking it away, and somebody needs to remember to tell them (eg AGENTS.md) to delete stuff.
I suppose there is a real demand for a harness that does get better over time for average users, i'm just a bit skeptical if that actually works out over a longer time.
My interest was piqued and I tried to read on… Holy mother of all that is slop, I couldn’t even get to the text because of all the interactive slop I had to scroll through.
This reminds me of yoyo-evolve a little bit https://github.com/yologdev/yoyo-evolve
I always found Erlang (and Elixir) to be an interesting use case for this kind of agentic harness. It can run distributed across machines and can hot swap any function. It's order of magnitudes faster to update, and often easier to introspect and keep compatible over time.
I built something for this and also found that the idea of a "guardian" core that can detect misbehavior (crashing etc) and roll back a messed up state really helped. But it's a lot of work to get a harness to behave well, and each model reacts differently to the harness limitations, so I found more success in building a meta-harness instead (just run existing harnesses inside a common UI and inject all capabilities).
Why recompile the harness? Why not just allow a flexible plugin system like maki [1], which takes its plugin philosophy from Neovim.
[1]: https://maki.sh/#lua
So people can use Rust, and not Lua?
Ok i should not be trying to wrap my head around this, this late at night - but i got to be frank and tell you this is way too loud and indecipherable. I need a two sentence explanation given that i work with `pi` and build my own tools.
Do i have it right that its some prompts injection for reflection and a composition scheme for the prompt/context where each element is version controlled?
If so, my experience with all the SOTA models is that they're pretty bad at finding the 'clean' solution to self improvements; eg they still always add text instead of taking it away, and somebody needs to remember to tell them (eg AGENTS.md) to delete stuff.
I suppose there is a real demand for a harness that does get better over time for average users, i'm just a bit skeptical if that actually works out over a longer time.
Might check it out later.
My interest was piqued and I tried to read on… Holy mother of all that is slop, I couldn’t even get to the text because of all the interactive slop I had to scroll through.
it may be slop, but its very pretty slop
[dead]