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.

10 pages 43 min 9,183 words

01 the section

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