Agent Engineering
A comprehensive resource for understanding AI agent architectures, design patterns, and implementation strategies. From basic loops to advanced multi-agent systems.
Code Examples
Every pattern includes implementations in pseudo-code, Python (LangChain/LangGraph), and C# (Microsoft Agent Framework).
function agentLoop(task):
context = initializeContext(task)
while not task.isComplete():
observation = perceive(context)
thought = reason(observation)
action = decide(thought)
result = execute(action)
context = update(context, result)
return context.result def agent_loop(task: Task) -> Result:
context = initialize_context(task)
while not task.is_complete():
observation = perceive(context)
thought = reason(observation)
action = decide(thought)
result = execute(action)
context = update(context, result)
return context.result public Result AgentLoop(Task task)
{
var context = InitializeContext(task);
while (!task.IsComplete())
{
var observation = Perceive(context);
var thought = Reason(observation);
var action = Decide(thought);
var result = Execute(action);
context = Update(context, result);
}
return context.Result;
} Topics
Core concepts and patterns for building AI agents.
Tool Use & Function Calling
Learn how agents extend their capabilities by calling external tools, APIs, and databases.
ReAct Pattern
Master the Reasoning + Acting loop that powers agentic AI behavior.
Memory Systems
Explore how agents maintain context and learn from past interactions.
Context Engineering
Optimize what goes into the context window. Write, Select, Compress, and Isolate strategies.
Context Bloat & Rot
Understanding and mitigating performance degradation in large contexts.
Prompt Caching
Reduce costs and latency by caching attention computations across requests.
Skills Pattern
Filesystem-based tool management with 98% token savings.
Model Context Protocol
Open protocol for connecting agents with tools and data sources.
Agent2Agent Protocol
Protocol for agent-to-agent discovery and collaboration.
Learning & Adaptation
Token-space learning, reflexion, and self-evolving agents.
Multi-Agent Systems
Discover patterns for coordinating multiple specialized agents.
Agentic RAG
Beyond simple retrieval: Self-RAG, Corrective RAG, and Graph RAG.
Safety & Guardrails
Implement safeguards to ensure agents behave safely and predictably.
Evaluation & Metrics
Measure agent performance: benchmarks, metrics, and LLM-as-Judge.
Benchmarks
Industry-standard benchmarks for measuring agent capabilities.
- SWE-bench
- Software engineering task completion
- WebArena
- Web browsing and interaction tasks
- GAIA
- General AI assistant capabilities
Top agent scores as of 2025
Resources
Curated papers, frameworks, and tools for building AI agents.