Beating AI News Flash: TypeSafe founder Diogo Almeida has publicly released a Coding Agent design draft, hoping the community will take it and experiment with it directly. He stated plainly that his team most likely doesn't have time to build all these ideas themselves.The most core change in this draft is context. Current Coding Agents usually carry previous chats, tool results, and code state throughout, and only compress or restart once it gets too long. TypeSafe wants the Agent to reassess each round: whether to continue using the existing cache or reselect a more suitable context; what to keep in full text, what to keep only as a summary, and what to discard entirely.Almeida believes that many existing Agent designs are in turn constrained by KV Cache. For example, if a task is first handed to a cheap model and then switched back to a strong model, the strong model may still have to reread the long context, ultimately making it more expensive; tool definitions stuffed into the system prompt for the long term also permanently take up space.If context can be dynamically reorganized, model routing, Sub-agent, MCP, and AGENTS.md can also be loaded on demand accordingly. Simple tasks go to cheap models, tools are loaded only when needed, and sub-agents receive only the state relevant to their own tasks. Almeida tentatively calls this approach "Meta-attention."