Running Astra agents overnight lets the workspace grow into a tree on its own
Without designing the structure in advance, the author let multiple agents run repeatedly, and research, decisions and past attempts settled into a reusable tree-shaped workspace.
i left GPT-6 Astra running overnight and came back to something I didn’t expect the workflow had started as a mess different agents were researching the same things, reopening old questions and rebuilding context that already existed somewhere else every run worked but every run felt like the first one then the system slowly started looking different one piece of research connected to another old decisions stayed attached to the work that created them useful paths kept getting deeper while dead ends stopped expanding after enough runs, the workspace started forming the structure you see in the video that was the weird part I hadn’t built the tree first and forced the agents into it the tree was what remained after the agents kept working new tasks could now land inside something that already had history an agent could pick up a problem and find the research, decisions and previous attempts sitting around it instead of reconstructing everything from scratch the model was still GPT-6 Astra the tasks were still the same but the starting point had changed completely the first few runs built the work the runs after that started building on it
This post quotes another post: x.com/lummox_eth/status/2100288151897981255