TypeSafe founder Diogo Almeida has published a design draft for a Coding Agent and said the community is free to experiment with it directly. He also said the team likely will not have enough time to build every idea in the proposal themselves. The main change in the draft is how context is handled. Instead of keeping prior chat history, tool outputs, and code state attached until the context becomes too long and must be compressed or reset, the proposal suggests that the agent should decide again on every turn what it actually needs to keep. That includes whether to reuse the current cache or assemble a better context set, which items should remain in full, which should be reduced to summaries, and which should be dropped entirely. Almeida argued that many current agent designs are being constrained by KV Cache in reverse. He gave examples such as routing a task to a cheaper model first and then back to a stronger model, only for the stronger model to reread a long context and erase the expected cost savings. He said the same issue applies when tool definitions stay in the system prompt for long periods and keep consuming space. Almeida is currently calling the approach "Meta-attention."
TypeSafe founder Diogo Almeida has released a design draft for a Coding Agent and said he wants the community to take it and experiment with it directly. He added that the team will most likely not have time to build all of the ideas on its own.
Context handling is the main change
The core change in the draft is context. Almeida said current Coding Agents usually keep carrying prior chat history, tool results, and code state, then compress or restart once that context grows too large.
TypeSafe's proposal is to let the agent make that decision again on every turn: whether to keep using the existing cache or assemble a more suitable context set, which parts should stay in full, which should be kept only as summaries, and which should be discarded.
Almeida says KV Cache is shaping agent design the wrong way
Almeida argued that many existing agent designs are being constrained by KV Cache in reverse. In one example, a task may first be routed to a cheaper model and then sent back to a stronger one, but the stronger model may need to reread a long context, making the overall process more expensive instead of cheaper.
He also said tool definitions that remain in the system prompt for long periods keep taking up context space.
Models, tools, and sub-agents could load only when needed
If context can be rebuilt dynamically, then model routing, sub-agents, MCP, and AGENTS.md can also be loaded on demand. Simpler tasks can go to cheaper models, tools can be loaded only when needed, and each sub-agent would receive only the state tied to its own task.
For now, Almeida is calling the approach "Meta-attention."
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
