AI agents

Building with language models that use tools: what makes an agent, how its context is managed, retrieval, patterns for several agents, and keeping one safe, tested and observable. The code examples are in Python.

Foundations

AGENTS

Agents from first principles

still growing

Explain what makes a model an agent, and build the core loops of tool use, planning, ReAct and reflection.

0/9 read
9 notes
  1. Why agentic patterns
  2. LLM versus agent
  3. Agent design principles
  4. State graphs
  5. Agent memory
  6. Tool use
  7. Planning
  8. ReAct
  9. Reflection
AGCTX

Context engineering

still growing

Decide what goes into an agent's context window on every call, and trim, order and scope it when it grows.

0/11 read

Builds on: Agents from first principles

11 notes
  1. Prompt engineering
  2. Context engineering
  3. Anatomy of context
  4. The context assembler
  5. Context overflow
  6. Context compression
  7. Context reranking
  8. Context scoping
  9. Flow engineering
  10. Graph engineering
  11. The dynamic expert agent
AGRAG

Retrieval-augmented generation

still growing

Explain how RAG puts fresh and private data in front of a model, and choose between RAG, long context and agentic retrieval.

0/8 read

Builds on: Context engineering

8 notes
  1. Why RAG
  2. Embeddings
  3. Vector databases
  4. RAG ingestion
  5. RAG at query time
  6. RAG versus long context
  7. Agentic RAG
  8. Context reranking

Several agents

AGPAT

Multi-agent patterns

still growing

Choose between a router, a handoff, a group chat and a swarm, and say what each one costs.

0/6 read

Builds on: Agents from first principles

6 notes
  1. The monolith agent problem
  2. Router pattern
  3. Handoff pattern
  4. Group chat pattern
  5. Swarm and map-reduce
  6. Choosing a pattern
AGPROTO

Agent protocols

still growing

Say which protocol connects an agent to a tool, to another agent, and to a user, and what one MCP tool call looks like on the wire.

0/4 read

Builds on: Agents from first principles, Multi-agent patterns

4 notes
  1. Why agent protocols
  2. MCP
  3. MCP lifecycle
  4. A2A, ACP and AG-UI

Safety and operations

AGSAFE

Safety and guardrails for agents

still growing

Name the risks of an agent that reads text and calls tools, and put the right control in front of each one.

0/7 read

Builds on: Agents from first principles

7 notes
  1. Probabilistic risks
  2. LLM guardrails
  3. Hallucination prevention
  4. Human in the loop
  5. Ejection logic
  6. Secure tooling
  7. Agent ethics
AGOPS

Evaluating and running agents

still growing

Test an agent before release, watch it after, and follow the lifecycle that connects the two.

0/7 read

Builds on: Agents from first principles, Safety and guardrails for agents

7 notes
  1. Agent evaluation
  2. Test datasets and mock tools
  3. LLM as judge
  4. Agent ops
  5. Traceability
  6. Agent metrics
  7. The agent development lifecycle

All topics →