Shrutam CCA-F Crash › Domain 3 › d3-l05-isolated-subagent Hinglish English →

Subagent Dispatch — Isolated Context Bubbles

Domain 3 · 20% ~12 min Hinglish narration

Audio-only (commute / mobile data)

Same Saavi narration, smaller file. Opus 48k preferred — auto-selected by your browser.

Scenario anchor

Aaranya IT ke ek BFSI client ke liye aap ek large-scale codebase migration orchestrate kar rahe hain — 40+ microservices, har ek ka apna repo, apna schema, apna test suite. Agar aap sab kuch ek hi Claude session mein push karein, toh context window overflow aur cross-service token contamination — dono guaranteed hain. Yeh lesson precisely yahi resolve karta hai: subagent dispatch through isolated context bubbles, jisme har subagent apne API-gateway microservice ki tarah independently operate karta hai, bina upstream state leakage ke.

Key Takeaways

Each subagent receives only the minimal, scoped context it needs — treat it like a stateless microservice invocation, not a shared-memory thread.
Parallel subagent dispatch is the primary throughput lever at code-at-scale; isolation is not optional hygiene, it is the architectural guarantee that makes parallelism safe.
The orchestrator holds canonical state; subagents return structured results (JSON schema-validated) and are never allowed to write back into each other's context bubbles.
Memory anchor: subagent isolation maps directly to API-gateway microservices — each call is independently authorized, scoped, and stateless; the gateway (orchestrator) owns routing and aggregation.

← Previous lesson Same lesson in English Next lesson → Back to overview