Evaluating AGENTS.md: Are Repository-Level Context Files Helpful for Coding Agents? finds that repo-level AGENTS.md files often reduce coding agents’ task success and raise inference cost, arguing for minimal, targeted context. Outcome engineers should treat context size and placement as an engineering variable — lean, task-specific context beats bulk repo dumps for correctness and cost (Principle 06, 16).

Temporal raises $300M to scale AI agent reliability platform reports a major funding round to expand a cloud platform focused on AI agent reliability. Outcome engineers gain a production-grade model for orchestrating retries, state, and observability across autonomous agents — adopt platform-level reliability primitives to ship resilient agent-backed features (Principle 14).

The Multi-Model Database for AI Agents: Deploy SurrealDB with Docker Extension shows SurrealDB packaging vectors, graphs, documents, and relational data into a single low-latency engine that simplifies agent memory and RAG pipelines. That shifts agent architecture: a unified context store reduces glue code and latency, making memory, graph reasoning, and retrieval engineering more legible (Principle 06, 11).

Anthropic launches Claude Sonnet 4.6 with coding and consistency improvements, plus 1M-token context window in beta announces coding boosts, consistency improvements, and a beta 1M-token context window. Longer in-model context changes trade-offs for retrieval, orchestration, and prompt engineering—rethink when to store state externally versus keep it in the model (Principle 06).

A Guide to Which AI to Use in the Agentic Era argues that model behavior depends as much on harnesses and deployment patterns as on raw model choice. For outcome engineers, evaluate models inside your harness early—tooling, orchestration, and runtime wrappers often determine agent reliability and outcomes more than benchmark numbers (Principle 09, 06).