Agents at Scale — 1M Context, Agentic Retrieval, and Context Ops
1M context is now generally available for Opus 4.6 and Sonnet 4.6. Anthropic makes 1M-token context generally available for Opus 4.6 and Sonnet 4.6 at standard pricing. Outcome engineers can now design agents that hold whole documents, multi-file state, and long histories without stitching hacks — this shifts context-engineering tradeoffs and reduces reliance on external memory systems (Principles 06,12).
Beyond Semantic Similarity: Introducing NVIDIA NeMo Retriever’s Generalizable Agentic Retrieval Pipeline. NVIDIA releases NeMo Retriever, an agentic retrieval pipeline that loops LLM reasoning with retrievers to generalize across retrieval benchmarks and win leaderboard spots. Outcome engineers get a practical pattern for integrating reasoning+retrieval loops to improve grounding and retrieval generalization, a direct lever for agentic coordination and legible retrieval (Principles 06,09).
Context Gateway — Compress agent context before it hits the LLM. Compresr’s Context Gateway pre-computes and compresses agent history so conversations never stall when hitting LLM context limits. This gives practitioners a concrete off-ramp from brittle truncation: compress and surface selective context to keep agents coherent under token caps (Principles 06,11).
How Windsurf Cascade Actually Understands Your Codebase. Windsurf’s Cascade builds dependency graphs and AST models to drive selective context, enabling coherent multi-file refactors that are type-accurate within finite context windows. Outcome engineers can adopt program-structure-aware context extraction to feed agents minimal, precise code context for edits and reviews — a concrete implementation of legible landscapes and graph-driven context (Principles 06,11).
Optimizing Content for Agents. Sentry reworks docs and endpoints to serve agents Markdown and structured APIs, reducing context bloat and improving agent accuracy. This demonstrates that documentation and API design are first-class inputs for agents — outcome engineers should build content negotiation and structured doc surfaces to make agentic workflows reliable (Principles 06,13).