Context cost Speed efficiency

Feature request for product/service

AI Models

Describe the request

Dear Cursor,

I am a programmer developing applications, libraries, and games. While the platform already offers a variety of MCPs and skills, I find that it lacks the flexibility needed to break tasks into smaller agents instead of a single monolithic one.

The problem
In the current “agent” mode, a single agent handles everything. This is convenient but heavily burdens the model: it inflates context costs and reduces performance.
For large projects (multi‑page sites, complex games), a single agent quickly becomes overloaded with information.
My proposal: a multi‑agent architecture
Role segregation

A “smart” coordinator model handles only high‑level planning and task distribution; it does not get involved in low‑level implementation details.
For each concrete subtask, a specialized model (“simple” agent) receives just the relevant context.
Benefits

Reduced context size – each model sees only the information it truly needs.
Cost savings – fewer tokens are spent on storing and transferring data between agents.
Speed gains – parallel execution of several lightweight models is faster than a single heavy one.
Flexibility – new sub‑tasks can be added without reconfiguring the entire system.
Example use case: a multi‑page web site

The main agent drafts an overall plan and allocates tasks:
A “Database” module designs schemas and generates migrations.
An “API” module writes endpoints and business logic.
Each page gets its own agent responsible for HTML/CSS/JS.
After each sub‑agent completes its work, it reports back to the main agent, which verifies correctness and stitches everything into a final build.
Bottom line

Context + cost → decrease.

Speed + efficiency → increase.

If you’re interested in discussing implementation details (API interfaces between agents, context‑exchange mechanisms, training specialized models), I’d be happy to share my ideas and sample code.

Thank you for your consideration!