// verdict · checked 2026-08-11 · LangGraph OSS Python (docs.langchain.com, Aug 2026)

LangGraph

LangChain

LangGraph is a wrapper with ambition. It passes 2 of the five criteria in our AI agent test. This is a dated assessment of the shipped product, not a claim about every possible setup.

2/5
WRAPPER WITH AMBITION

Graphs all the way down. You bring the agent, it brings the arrows.

CAPABILITY
2/5 WRAPPER WITH AMBITION
PROOF
OPERATOR INSPECTABLEYou can run it and read the steps yourself — the record exists for whoever runs it
TRACE ACCESS
OPERATOR-OWNEDWhoever runs it holds the record; nothing is published
LEASH
Not measuredno execution record scored — measure it yourself
COMMUNITY
Loading votes…cast yours on the full list

Compare → Download the card Bring a trace →

LangGraph is infrastructure for building agents, not a shipped agent. It scores low because we score what a framework does out of the box - a category distinction, not a product failure.

01

The five questions

plain-English test · the rules →
LOOPSruns until doneno Framework rule: score default out-of-box behavior. LangGraph executes the developer-wired graph; cycles are edges the developer authors and recursion_limit is only a cap. The loop is a provided primitive, not the framework deciding step count. OPERATOR INSPECTABLE source · confidence 8.8/10 · checked 2026-08-11 · ruled by hand
CHOOSESpicks its own toolsyes Prebuilt create_react_agent/create_agent loops on tool_calls in the model's reply, so the model picks which tool and in what order at runtime; the developer supplies only the tool list. OPERATOR INSPECTABLE source · confidence 8/10 · checked 2026-08-11 · ruled by hand
ACTSchanges the worldno langgraph.prebuilt ships only scaffolding (ToolNode, create_react_agent, ValidationNode) and zero concrete tools, so every file write, HTTP call or command lives in developer-authored nodes. OPERATOR INSPECTABLE source · confidence 7/10 · checked 2026-08-11 · ruled by hand
RECOVERSfixes its own errorsno Fault-tolerance docs: retry_policy is opt-in per add_node and re-runs the IDENTICAL failed node; ToolNode's default only echoes bad-arg errors back, alternate routing is developer-authored. OPERATOR INSPECTABLE source · confidence 8/10 · checked 2026-08-11 · ruled by hand · upheld in court #0001
UNSUPERVISEDnobody watchingyes No default approval gate: HITL is opt-in middleware the developer adds to the agent's middleware list, and a compiled graph invoked without interrupt() runs to END unattended. OPERATOR INSPECTABLE source · confidence 8/10 · checked 2026-08-11 · ruled by hand

Capability decisions combine documented behaviour and execution evidence. The proof label on each row shows which of the two decided it, and Leash is reported separately, only when it was measured from an execution record. Primary source: docs.langchain.com. Tested against: LangGraph OSS Python (docs.langchain.com, Aug 2026). Our confidence in this dossier: 7/10.

02

Can you see it work?

the weakest of the five grades above

OPERATOR INSPECTABLE — You can run it and read the steps yourself — the record exists for whoever runs it. Trace access: OPERATOR-OWNED — Whoever runs it holds the record; nothing is published.

LangSmith traces every node and tool call; thread-history API returns the run

Where we looked: docs.langchain.com/llms.txt: LangSmith observability, view-traces, trace-with-api, distributed tracing; LangGraph Studio; thread-state, thread-history and list-runs APIs. · see the evidence →

  1. VERDICT UPHELD
    RECOVERS no → no filed by editorial internal audit · none — filed by the editors against their own verdict · 2026-08-11

    The claim was that RetryPolicy and error handlers constitute recovery. Rule 04 excludes repeating the identical call, and RetryPolicy re-runs the identical node attempt; the error handler is a developer-written hook, a primitive the framework provides rather than the framework devising an alternative. Verdict upheld.

    read the case →
04

Revision history

rss for this product
  1. 2026-08-11
    1/5 CRON JOB WITH EXTRA STEPS → 2/5 WRAPPER WITH AMBITIONPre-launch adjudication: every criterion the verdict depends on was re-checked by hand against the published rules.editorial correction — no case was filed
05

Badges

earned, never for sale

OPERATOR INSPECTABLE VERDICT UPHELD

These render from the live dataset, so a badge cannot outlive the verdict it claims. If the verdict changes the badge changes with it, and an unearned one returns 409 rather than an image. No sponsor can buy one, at any price.

06

Think this is wrong?

bring evidence, not opinion

Name the criterion and link a trace, a doc, or a recorded run that shows the behaviour. A verdict changed by evidence is the best thing that can happen to this site, and the change goes on the public record with your reason attached. Vendors are welcome, and a vendor's own filing is labelled as such rather than buried.