Context Engineering and Memory for AI Agents: Beyond Prompting

An advanced program for AI/ML engineers and software engineers building agents and LLM applications who want to move beyond prompting toward the systematic engineering of context and memory. You learn to treat the context window as a finite resource that must be budgeted and kept clean, to decompose the anatomy of context (system prompt, instructions, examples, conversational history, tool results, retrieved data) and to design memory layers for agents: working/short-term, semantic, episodic and procedural. Covers memory managers and persistence (extraction, consolidation, LangMem-style memory store patterns), retrieval strategies for context (vector vs graph vs relational and the hybrid approach), compaction and window management (summarization, pruning, sliding window, context offloading), context engineering for multi-step agents and multi-agent systems, plus reliability evaluation, cost and token economics, prompt caching and optimization. Everything anchored in official Anthropic, OpenAI and LangChain (LangMem and LangGraph) documentation and the DeepLearning.AI course, with 2026 models (Claude Opus 4.7/4.8, GPT-5.5, Gemini 3.1 Pro) and a capstone project: an end-to-end agent with persistent memory, evaluated for reliability. The technical content is informational and versions the APIs and models, which may change.