Along with the recent results on HRM/TRM, it seems like you can squeeze a lot more juice out of LLMs just by having some sort of iterative refinement?
I can buy that it works, but seems like the agent harness needs to be specific to the task at hand. It's hard to find more info on Poetiq but based on their site
>Our approach allows us to find effective task-specific reasoning strategies using much less data (hundreds of data points, rather than millions), while being compatible with the LLMs you’re already using.
I'm guessing it's something like automatic prompt-engineering + automatic agent topology exploration.
