This resonates. What’s being framed here as “new insight” around agents stopping conditions, human oversight, boundaries are exactly the questions we discussed earlier, just from a different angle.
The hard part isn’t building agents. It’s designing when they shouldn’t act, when a human needs to step in, and how responsibility stays visible over time. Those are not technical problems first; they’re UX and governance problems.
Good to see the technical discourse catching up to that layer.
Outstanding breakdown of the async generator patter here. Using `for await` to handle streaming is exactly how you avoid blocking whle managing tool execution loops. The permission modes as async callbacks is genius - clean architecture that acutally works.
Does the SDK require using an API key? 🤔
A few weeks ago, I was experimenting with opencode. I saw the little side panel widget showing the cost of usage - I thought they were estimates.
I was billed $120 - incremental charges by Anthropic, that hit my bank one by one by one. And this happened right after my 5x procharge had $100.
Lesson learned 🤣😖
This resonates. What’s being framed here as “new insight” around agents stopping conditions, human oversight, boundaries are exactly the questions we discussed earlier, just from a different angle.
The hard part isn’t building agents. It’s designing when they shouldn’t act, when a human needs to step in, and how responsibility stays visible over time. Those are not technical problems first; they’re UX and governance problems.
Good to see the technical discourse catching up to that layer.
Outstanding breakdown of the async generator patter here. Using `for await` to handle streaming is exactly how you avoid blocking whle managing tool execution loops. The permission modes as async callbacks is genius - clean architecture that acutally works.