LANGCHAIN PUB_DATE: 2026.05.04

LANGCHAIN SHIPS STREAMING V2 AND STANDARDIZES AGENTS ON CREATE_AGENT

LangChain changed how streaming works and is standardizing agent creation on create_agent. The latest [langchain-classic 1.0.5](https://github.com/langchain-ai...

LangChain ships streaming v2 and standardizes agents on create_agent

LangChain changed how streaming works and is standardizing agent creation on create_agent.

The latest langchain-classic 1.0.5 retargets deprecations to create_agent and bumps core versions.

Core now includes content-block-centric streaming (v2) per the same notes, which changes event shapes and parsing paths; plan for SSE/WebSocket updates.

langchain-anthropic 1.4.3 also lands a stability fix and aligns with the agent API shift. If you lean into reasoning and tools, expect higher test-time compute per this cost deep-dive.

[ WHY_IT_MATTERS ]
01.

Streaming v2 changes event structure, which can silently break parsers, UIs, and logging.

02.

create_agent consolidates agent setup, reducing API churn across providers and easing migrations.

[ WHAT_TO_TEST ]
  • terminal

    Run a minimal chain with streaming v2 and verify SSE/WebSocket parsing, backpressure, and partial updates across OpenAI and Anthropic.

  • terminal

    Migrate one production agent to create_agent and measure latency, token usage, and tool-calling behavior under load.

[ BROWNFIELD_PERSPECTIVE ]

Legacy codebase integration strategies...

  • 01.

    Search for deprecated agent constructors, enable deprecation warnings in CI, and gate rollouts behind a feature flag.

  • 02.

    Add telemetry for dropped events and malformed deltas; compare costs with and without reasoning turns.

[ GREENFIELD_PERSPECTIVE ]

Fresh architecture paradigms...

  • 01.

    Adopt create_agent and streaming v2 from day one to keep provider-neutral event handling.

  • 02.

    Model UIs and storage on content blocks instead of raw text deltas to simplify multi-model support.

Enjoying_this_story?

Get daily LANGCHAIN + SDLC updates.

  • Practical tactics you can ship tomorrow
  • Tooling, workflows, and architecture notes
  • One short email each weekday

FREE_FOREVER. TERMINATE_ANYTIME. View an example issue.

GET_DAILY_EMAIL
AI + SDLC // 5 MIN DAILY