Anthropic is building out their managed agents platform, adding Dreaming (memory) and Outcomes (rubrics). The idea I'm wrestling with: how d…

Anthropic is building out their managed agents platform, adding Dreaming (memory) and Outcomes (rubrics). The idea I'm wrestling with: how differentiated are these platform features really? I initially thought the model would "eat your scaffolding", an argument best made by @nicbstme. But then @hwchase17, @Vtrivedy10, and the @LangChain team have laid out strong cases for the value of open harnesses. What convinced me is you don't get visibility into how mechanisms are implemented under the hood, or the ability to control them. For instance, I might want to detect PII and prevent it from being stored as memory. How can I implement this with Dreams? @nicbstme counters that models will be post-trained for their harness. But I'm not sold on how much that actually matters. I don't think I buy building agents on frontier labs' platforms. 🤷🏿‍♂️

Source: Harrison Chase (X) | 2026-05-06