Frameworks and Competitors
Interviews often ask: "Would you use LangChain or the direct SDK?", "What is Semantic Kernel?", "Claude vs GPT?" Here are balanced answers.
25.1 Agent / LLM frameworks
| Framework | What it is | With Claude |
|---|---|---|
| Anthropic SDK + Tool Runner | Official, thin, direct API | Least abstraction, new features immediately |
| Claude Agent SDK | Claude Code's harness as a library | Best for files/bash/coding agents |
| LangChain / LangGraph | Python/JS framework: chains, tools, RAG; LangGraph = stateful graph workflows | langchain-anthropic package; many integrations |
| LlamaIndex | Focused on RAG / data indexing | Documents + retrieval pipelines |
| Semantic Kernel | Microsoft's .NET/Python/Java SDK: plugins, planners, memory | Claude through Microsoft.Extensions.AI |
| Microsoft.Extensions.AI | Common abstraction for .NET (IChatClient) | AsIChatClient() of the Anthropic C# SDK |
| AutoGen / Agent Framework | Microsoft's multi-agent conversation framework | Claude through a model client |
| Vercel AI SDK | TypeScript/React streaming UI | Anthropic provider |
25.2 Claude vs GPT vs Gemini: how to answer
| Point | What to say |
|---|---|
| No single "best" | Depends on the use case; decide by running evals on your own task |
| Claude's strengths | Coding and agentic tasks, long context (1M), instruction following, safety/honesty, the MCP + Claude Code ecosystem |
| What to compare | Quality on your evals, latency, cost per completed task, context size, tool use reliability, data/compliance, cloud availability |
| Multi-provider strategy | Abstraction layer (IChatClient / LangChain) + evals + a fallback provider |
| Image generation | Claude does not generate images (it only understands them); you need a separate model for that |