09 section
Frameworks & Tools
Ten frameworks judged on what they cost you in production — LangChain, LangGraph, LlamaIndex, DSPy, Semantic Kernel, CrewAI, AutoGen — ending in a decision matrix and the coding-agent landscape.
What is in here
One chapter per framework, each covering the abstraction it sells you and the failure mode it hides. Chapters 01 through 07 are independent, so read whichever stack you are evaluating, then read 08 — the decision matrix that ties them together. Chapters 09 and 10 cover autonomous coding agents, a different category entirely, and work best as a pair.
01
2 min
LangChain Deep Dive
LCEL's pipe operator, the split between core, community and partner packages, and where LangChain's abstraction layer starts costing you more than it saves.
frameworkorchestration
02
2 min
LangGraph Orchestration
Agents as graphs: nodes, conditional edges and a typed state schema, with checkpointing that lets a run pause for human approval and resume later.
frameworkorchestrationagents
03
2 min
LangSmith Observability
Trajectory tracing, gold-standard datasets and LLM-as-judge evaluators: the LLMOps loop that turns user complaints into regression tests instead of anecdotes.
frameworkobservabilityevaluation
04
3 min
LlamaIndex
The data-first framework: nodes carrying parent-child metadata, retrievers well beyond vector search, and the event-driven Workflows API that answers LangGraph.
frameworkragretrieval
05
3 min
DSPy
Replacing hand-tuned prompts with signatures and an optimizer: declare the task, define a metric, let MIPROv2 compile prompts that survive a model swap.
frameworkpatternsevaluation
06
2 min
Semantic Kernel
Microsoft's orchestration engine for C#/.NET shops: kernel functions, plugins, hierarchical planners, and Entra ID integration that survives enterprise security review.
frameworkorchestrationsecurity
07
3 min
AutoGen and CrewAI
Two rewrites compared: AutoGen 0.4's async AgentChat API against CrewAI's role-based crews and the newer Flows state machine, plus where each hides failures.
frameworkagentsorchestration
08
4 min
Framework Selection Guide
A decision matrix mapping requirement to stack: pure RAG to LlamaIndex, stateful human-in-the-loop to LangGraph, cross-model reliability to DSPy, plus the build-versus-buy line.
frameworkpatternscost
09
10 min
Claude Code
Anthropic's terminal-native coding agent end to end: the tool loop, the CLAUDE.md manifest, sub-agent parallelism, permission model, and headless use inside CI.
coding-agentsagentsframework
10
12 min
AI Coding Agents Landscape
The three-layer coding stack of open-weight models, AI-native IDEs and autonomous agents, benchmarked on SWE-bench with per-seat costs and a selection guide.
coding-agentsframeworkcost