Treat filtering as one layer and design for the case where prompt injection succeeds. The "lethal trifecta" as a design check, architecture patterns that separate reading untrusted content from taking actions, and the controls that limit damage when they fail.
Read ArticleTreat the context window like memory in a constrained system: allocate it, measure it and reclaim it. A per-component budget, a compaction policy that runs in code, just-in-time retrieval, and when to hand work to a sub-agent with a clean context.
Read ArticleOnce several services call several model providers, put a gateway between them. What belongs in it (credentials, budgets, logging, failover), what doesn't (prompt logic), and the two features that quietly change model behavior: cross-model failover and semantic caching.
Read ArticleAn agent that works for hours needs a recovery plan for context loss, process failure and drift. The harness pieces that make long runs recoverable: durable execution, a progress file the agent maintains, small verifiable increments, and budgets enforced from outside.
Read ArticleMost "agent" projects need a fixed workflow with one or two model calls, not an autonomous loop. How to tell which you need, the five workflow patterns worth knowing, and what changes in cost, testing and permissions when you do choose an agent.
Read Article