Part 4

Engineering Patterns for Reliable Agents

22 notesOpen reading · no login

Part 422notes

THE QUESTION THIS PAGE ANSWERS

ANSWER FIRST

What will the “Engineering Patterns for Reliable Agents” AI learning path help you do?

Read production-oriented patterns for context assembly, tool design, evaluation, long-running work, and sandboxing. The emphasis is on boundaries you can inspect when an agent succeeds for the wrong reason. The path contains 22 free notes, each centered on one question you can understand and test.

DECISION RULE

Core themes include Agent Design Patterns, The Art of Tool Design, Agent Evaluation, Long-Running Agents.

TRY NEXT

Begin with “Workflow vs Agent: Know What You Want First,” then choose the next note by the task in front of you.

WATCH FOR

Do not optimize for finishing the list. Explaining one trade-off with your own example matters more than opening more titles.

What this route helps you practice

Open the first note

Each chapter follows a class of real decisions. Follow the sequence, or enter at the problem you are solving today.

22notes

01Workflow vs Agent: Know What You Want FirstPredefined flows vs model-driven decisions — Anthropic's two categories of Agent systemsDesign Pattern4 min↗02Five Workflow PatternsPrompt Chaining / Routing / Parallelization / Orchestrator-Workers / Evaluator-OptimizerDesign Pattern5 min↗03From Prompt Engineering to Context EngineeringCurating the optimal Token combination for each inference round — prompt writing is just one pieceMethodology5 min↗04Three Context TacticsCompaction, structured notes, sub-Agent architecture — three context management strategies for long tasksMethodology6 min↗05ACI: Agent-Computer InterfaceTools are the contract between Agents and the world. Design Agent interfaces like you design human interfacesDesign Pattern7 min↗06Milvus as an Agent Knowledge ToolWrap vector retrieval as search_knowledge: ToolMessage, memory separation, and testing both calls and non-callsHands-on3 min↗07Think Tool: Making AI Think Before ActingGive the Agent a pause-to-think space in complex tool chains — τ-bench performance up 54%Deep Dive6 min↗08Using Agents to Optimize Agent ToolsClaude Code in practice: using AI to write tool descriptions, run evals, and auto-iterateHands-on6 min↗09Why Evaluation Matters More Than TrainingWithout evals, fixing one bug creates three. Anthropic's Eval methodologyMethodology6 min↗10Three Graders: Code, Model, HumanStatic assertions vs LLM-as-Judge vs human calibration — which fits which scenarioHands-on5 min↗11Eval Pitfalls: Noise, Cheating & RegressionInfra noise causes 6pp errors, models recognize tests, Prompt changes may drop Eval by 3%Case Study5 min↗12Why Agents Can't Handle Long TasksTrying to do too much at once, or quitting after one round — two classic failure modesCase Study4 min↗13Initializer + Coding AgentThe initializer sets up the environment, the coding Agent progresses incrementally — a dual-role Harness designDesign Pattern4 min↗14Managed Agent: Brain-Hand SeparationSplit thinking and execution into different processes — virtualizing Agents like an operating systemArchitecture5 min↗15Session ≠ Context WindowSession logs are persistent event streams; the context window is temporary working memory — they must be separatedDeep Dive5 min↗16Three Risk Categories: Abuse, Misbehavior, External AttackAnthropic's safety taxonomy: user abuse / model misbehavior / Prompt InjectionSecurity5 min↗17Sandbox & Credential IsolationGenerated code and secrets never share the same container. Structural security beats prompt-based securitySecurity5 min↗18Contextual Retrieval: Better RAGAdding context to chunks before retrieval — Anthropic's upgraded RAG approachDeep Dive5 min↗19Advanced OverviewDesign patterns, tools, evaluation, long-running, security — all connected in one diagramRecap4 min↗20Do the Simplest Thing That WorksAnthropic's core engineering philosophy: "Do the simplest thing that works"Finale4 min↗21Build Your First Eval SetBefore/after score demo; three tasks: collect ten real cases, write pass criteria and run a baseline, let the numbers talkHands-on6 min↗22AI Engineering Design Patterns · 30 Tough QuestionsEach with intent, framework, and bonus points: context engineering / long tasks / grep vs RAG / ACI tool design / eval infrastructure / LLM-as-Judge / sandbox isolationQuiz18 min↗