Building Effective AI Agents
Anthropic, 2024
The clearest writeup of when to reach for a workflow versus an agent. The whole DreamerOS engine-selection layer is downstream of this distinction.
(link to be added)
Library
The DreamerOS Reading Room
The papers, repos, and guides we cite when we build.
01
Papers
The papers we point teams to when they ask how the constellation actually works.
Agents Companion
Google, 2024
The complement piece. Patterns for tool use, memory, and multi-agent coordination at production scale.
(link to be added)
ReAct: Synergizing Reasoning and Acting in Language Models
Yao et al., 2022
The pattern most modern agents are still running. Reason about the next action, take it, observe, repeat.
(link to be added)
Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Wei et al., 2022
Why "think step by step" started showing up everywhere. The paper that made reasoning a first-class prompt structure.
(link to be added)
Tree of Thoughts: Deliberate Problem Solving with LLMs
Yao et al., 2023
Branching beats linear when the problem has more than one promising path. The intellectual ancestor of Mind-Eye depth calibration.
(link to be added)
Reflexion: Language Agents with Verbal Reinforcement Learning
Shinn et al., 2023
Agents that reflect on their own failures and update. The core idea inside IFP-Verify and the receipts log.
(link to be added)
Retrieval-Augmented Generation Survey
Gao et al., 2023
The honest tour of RAG. What works, what does not, where most teams over-engineer it.
(link to be added)
02
Repos
The code we read first when we want to learn how a piece of the stack actually behaves.
Anthropic Claude API examples
anthropics/anthropic-cookbook
Working code for tool use, prompt caching, batch, citations, and vision. Read before you build.
(link to be added)
OpenAI Cookbook
openai/openai-cookbook
The reference patterns for function calling, embeddings, and assistants. Useful even if you do not ship on OpenAI.
(link to be added)
LangChain
langchain-ai/langchain
The framework you should read to understand the abstractions, then decide whether you want them.
(link to be added)
LlamaIndex
run-llama/llama_index
The retrieval-first cousin to LangChain. Strong on data connectors and indexing primitives.
(link to be added)
MCP servers reference
modelcontextprotocol/servers
Reference implementations for Model Context Protocol servers. The plumbing that makes tool use portable across engines.
(link to be added)
03
Guides
The long-form writeups we hand to teams who want one document to read this weekend.
Building Effective Agents
Anthropic engineering blog
The companion essay to the agents paper. The framing we keep coming back to.
(link to be added)
Claude Code Best Practices
Anthropic
How to use Claude Code without wasting tokens on the wrong things. The CLAUDE.md and skills pattern starts here.
(link to be added)
A Practical Guide to Building Agents
OpenAI
The other side of the same coin. Good on evaluation, less opinionated than Anthropic's piece.
(link to be added)
Agent Whitepaper
Google
The most enterprise-shaped of the three. Strongest on planner-executor architectures and orchestration.
(link to be added)
DreamerAI Methodology
DreamerAI, internal
The six lenses we put on every engagement. Reverse Math, Sideways Thinking, Triple-I, Anti-Slop Check, Triangulation, Fingerprint Portability.
04
Books
The longer reads. Worth a flight, worth a weekend.
Understanding Deep Learning
Simon J.D. Prince
The textbook to read if you want to stop nodding at words like transformer and start understanding them.
(link to be added)
AI Engineering
Chip Huyen
The book on building production AI systems. Evaluation, deployment, monitoring, the boring parts that decide whether you ship.
(link to be added)
Designing Machine Learning Systems
Chip Huyen
The predecessor. Same author, same discipline, more on data and training pipelines.
(link to be added)
05
Videos and Talks
The talks worth the hour. Bring coffee.
Stanford CS25: Transformers and Agents
Stanford University lecture series
The graduate seminar that keeps inviting the people who built the systems to come explain them.
(link to be added)
Anthropic interpretability talks
Anthropic research
The clearest public window into how a frontier lab thinks about what is going on inside the model.
(link to be added)
Intro to Large Language Models
Andrej Karpathy
The one-hour talk we send to every new client. Still the best plain-English explainer of the stack.
(link to be added)