Framework vs Harness: How to Choose Between LangGraph, Deep Agents and Claude Code
Saboo_Shubham_ · x · 2026-10-31
A clear breakdown of the framework vs harness distinction in AI agent development:
- Frameworks (ADK, LangGraph, CrewAI) are libraries you import: they give you agent classes, tool interfaces and model adapters, while you decide how the loop runs, what goes into context, and how the agent checks its own work.
- Harnesses (Claude Code, Codex, Deep Agents, Hermes) are working agents out of the box, with loop, tools, context, memory and subagents built in.
The line is blurring: LangChain, once a framework company, now ships its own harness (Deep Agents), while the Claude Code harness can be imported like a library via the Claude Agent SDK.
Selection advice: pick a harness when your value is the domain, data or workflow; pick a framework when the loop itself is your product or you need determinism and auditability; mix both when needed.
More from coding & agent
- Google Antigravity team teases wave of launches next week — prateeky2806 · 2026-10-31
- Running AI agents on a real Android phone: what broke first — Orfeo_Lyrakis · 2026-10-31
- A Budget That Resets on Every Retry Isn't a Budget: Scope Agent Limits to the Task — Key-Register-1969 · 2026-10-31
- CortexDB Packs Knowledge Graph, Agent Memory, and MCP Into a Single SQLite File — JeremyCMorgan · 2026-10-31
- Cloakwall: Stdlib-Only PII Redaction With Tamper-Evident Audit Logs for LLM Traffic — OldBodybuilder7090 · 2026-10-31
- Claude finds a real 0-day in project dependencies from a single prompt — DanielLockyer · 2026-10-31