LCEL vs legacy chains: what current LangChain code looks like

TL;DR: Current LangChain composes runnables with the pipe operator (LCEL). LLMChain and the other legacy chains moved to langchain-classic in v1. Chains are for straight pipelines; the moment the model needs to decide, that is an agent.

The fix

1. For new code, use LCEL (prompt | model | parser). Do not reach for LLMChain in new code; it lives in langchain-classic now and only exists for old codebases.
2. LCEL runnables get .invoke, .batch, .stream, .astream_events, .with_retry, and .with_fallbacks for free. If you are hand-rolling retry loops around a chain, you are reimplementing built-ins.
3. Prefer create_agent over chains for anything with tools. Chains are for straight pipelines; the moment the model needs to decide, that is an agent.

When to use this

• This covers exactly what the title says: LCEL vs legacy chains.
• You are setting this up for the first time, or auditing an existing setup.
• You want the key gotchas in one place before you start.

When not to use this

• You are doing a different workflow with LCEL; these steps are specific to the title above.
• You need the full reference docs; this is the short path, not the manual.

Compatibility

• Mentions 5.4 in the original; behavior may differ on much older or newer releases.

Verified against the official LangChain docs on 2026-09-26. Migration guide: https://docs.langchain.com/oss/python/migrate/langchain-v1