Curated Resources
These are the resources that matter. Not a list — a curated library. Every resource here was selected because it changed how experts think about AI engineering.
Foundational Papers
Context Engineering
| Paper | Authors | Why It Matters |
|---|---|---|
| Attention Is All You Need | Vaswani et al. (2017) | The transformer architecture that underlies everything |
| Lost in the Middle: How Language Models Use Long Contexts | Liu et al. (2023) | The evidence behind the "critical info at top/bottom" rule |
| REALM: Retrieval-Augmented Language Model Pre-Training | Guu et al. (2020) | Foundational RAG paper |
| Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks | Lewis et al. (2020) | The canonical RAG paper |
Agent Engineering
| Paper | Authors | Why It Matters |
|---|---|---|
| ReAct: Synergizing Reasoning and Acting | Yao et al. (2022) | The ReAct pattern for tool-using agents |
| Tree of Thoughts | Yao et al. (2023) | Multi-path reasoning for complex problems |
| Chain of Thought Prompting Elicits Reasoning | Wei et al. (2022) | Evidence for explicit reasoning steps |
| Reflexion: Language Agents with Verbal Reinforcement Learning | Shinn et al. (2023) | Reflection and self-critique patterns |
| Plan-and-Solve Prompting | Wang et al. (2023) | Plan before execute for complex tasks |
Evaluation
| Paper | Authors | Why It Matters |
|---|---|---|
| RAGAS: Automated Evaluation of Retrieval Augmented Generation | Es et al. (2023) | Foundation for RAG evaluation |
| Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena | Zheng et al. (2023) | LLM-as-judge biases and calibration |
| Large Language Models are not Fair Evaluators | Wang et al. (2023) | Position bias in LLM evaluation |
Security
| Paper | Authors | Why It Matters |
|---|---|---|
| Prompt Injection Attacks Against LLM-Integrated Applications | Greshake et al. (2023) | The first systematic study of prompt injection |
| Not What You've Signed Up For: Compromising Real-World LLM-Integrated Applications | Greshake et al. (2023) | Indirect prompt injection via tool outputs |
Books
| Book | Authors | Why Read It |
|---|---|---|
| Designing Machine Learning Systems | Chip Huyen | Systems thinking for ML in production |
| Building LLM Apps | Various | Practical LLM application patterns |
| The Pragmatic Programmer | Hunt & Thomas | Still the best engineering mindset book |
| Release It! | Michael Nygard | Stability patterns for production systems |
| Accelerate | Forsgren, Humble, Kim | Evidence-based DevOps — applies directly to AI DevOps |
Essential GitHub Repositories
| Repository | What It Is | Why Follow It |
|---|---|---|
| anthropics/anthropic-cookbook | Claude usage examples | Official Anthropic patterns |
| openai/openai-cookbook | OpenAI usage examples | Official OpenAI patterns |
| langchain-ai/langchain | LLM application framework | Most widely used LLM framework |
| langchain-ai/langgraph | Stateful agent graphs | Production agent orchestration |
| deepeval-ai/deepeval | LLM evaluation framework | CI/CD evaluation gates |
| explodinggradients/ragas | RAG evaluation | RAG quality measurement |
| promptfoo/promptfoo | Prompt testing | Security and comparison testing |
| microsoft/autogen | Multi-agent framework | Microsoft's agent orchestration |
| modelcontextprotocol/servers | MCP server library | Reference MCP implementations |
Industry Blogs Worth Following
| Blog | Organization | Focus |
|---|---|---|
| Anthropic Research | Anthropic | Safety, alignment, Claude capabilities |
| OpenAI Blog | OpenAI | GPT, agents, DALL-E, safety |
| Google DeepMind Blog | Google DeepMind | Gemini, research breakthroughs |
| The Batch | DeepLearning.AI | Weekly AI news digest |
| Chip Huyen's Blog | Chip Huyen | Practical ML engineering |
| Lil'Log | Lilian Weng (OpenAI) | Deep technical explanations |
| Eugene Yan's Blog | Eugene Yan | Applied ML and LLM systems |
| Simon Willison's Blog | Simon Willison | LLM security, tools, agents |
Standards and Specifications
| Standard | Organization | Relevance |
|---|---|---|
| EU AI Act | European Union | Mandatory compliance for EU deployments |
| NIST AI Risk Management Framework | NIST | US AI risk framework |
| Model Context Protocol Spec | Anthropic | MCP implementation standard |
| OpenTelemetry Specification | CNCF | Observability standard |
| Semantic Versioning | Community | Version management for prompts and agents |
Courses
| Course | Platform | Who It's For |
|---|---|---|
| Short Courses by DeepLearning.AI | DeepLearning.AI | Practical AI engineering, updated frequently |
| Building Systems with the ChatGPT API | DeepLearning.AI | Production LLM systems |
| LangChain for LLM Application Development | DeepLearning.AI | LangChain fundamentals |
| Building and Evaluating Advanced RAG | DeepLearning.AI | Production RAG systems |
Technology Radar
Updated quarterly. Last updated: Q3 2026.
Adopt (Use in production now)
- LangGraph — Stateful agent orchestration
- DeepEval — LLM evaluation in CI/CD
- RAGAS — RAG evaluation
- Langfuse — LLM observability
- MCP — Agent-to-tool integration standard
- PydanticAI — Type-safe LLM interactions
Trial (Evaluate for your use case)
- OpenAI Agents SDK — Native agent orchestration
- Google A2A Protocol — Agent-to-agent communication
- Braintrust — Prompt management and evaluation
Assess (Watch, not yet ready for production)
- Knowledge Graphs + LLMs — GraphRAG patterns maturing
- Mixture of Agents — Multiple models collaborating
- Long context models — 1M+ token windows with reliable recall
Hold (Do not adopt)
- Autonomous agents without human-in-loop for high-stakes tasks — Not ready for production
- LLMs for real-time decisions in regulated industries without human review — Compliance risk
Version: AIES v1.0.0✏️ Edit this page on GitHub