LangChain

You will like this if you want an agent by lunch: create_agent takes a model, your tools and a prompt, and the daily trait is that the model is a string, so moving from GPT to Claude to a Gemini flash model is a one-line change and the rest of the code stays put. That freedom is worth more than any single feature once the invoices arrive.

You will not like the size. Choosing between the base library, its graph runtime and Deep Agents is a decision before the first line. Pick it for breadth. Pick OpenAI Agents SDK if one vendor is fine and you want less to read.

The risk is the dependency matrix. The library is three layers: langchain-core, langchain, and a provider package per vendor pulled in as an extra such as langchain[anthropic]. Every one of those moves on its own schedule. A lock file that worked in spring resolves differently in autumn, and the error surfaces inside a tool call rather than at import.

The consequence: pin every layer and treat provider packages as part of your own release. What it does right is create_agent's small signature. The harness is minimal by design, so when something breaks, the surface you are debugging is the middleware you wrote, not a thousand lines of chain.

The design is a thin harness on a thick runtime. 1. The agent loop runs on LangGraph, which supplies persistence, durable execution and human-in-the-loop as inherited properties rather than features of the harness. 2. Tools are plain functions whose docstrings become the schema, so the model's view of a tool is whatever the author wrote. 3. Middleware wraps the loop for guardrails, retries, routing and per-tool policies.

Item 3 is where verification lives, which means it lives wherever the developer puts it. Nothing is verified by default. The documentation is complete on mechanics and silent on measured outcomes; there is no benchmark to audit, which is at least an honest silence.

This is the open-core playbook executed properly. The library costs nothing and produces traces; LangSmith meters them at 5k a month on the free tier and 10k on Plus, then pay-as-you-go, and hosted deployment is billed in compute units at $1.50 per LCU. The funnel is the install base, and the install base is the largest in the category.

The moat is not the code, which is MIT and forkable; it is the habit. Teams who trace in LangSmith do not leave over a price change. Likely acquirer: a hyperscaler that wants the funnel, or an observability vendor that wants the traces. Position: long the company, longer the library.

The demo is an agent in a notebook. Procurement is about the platform behind it: LangSmith Plus is $39 a seat, $2,340 a month for sixty, and the sign-on on that tier is Google and GitHub login, not SAML. Custom SSO, RBAC and self-hosted deployment are Enterprise, priced by conversation. Traces contain prompts and prompts contain customer data, so the retention question goes to legal before the seats go to finance.

CI fit is fine, it is Python. Onboarding is a week of reading for a mid-level engineer because the ecosystem is wide. Approved with conditions: library now, platform after the retention review.

MIT, and forked 24 thousand times, so the question of whether it survives the company was answered years ago. My local box goes in through the Ollama provider, HuggingFace is there for the odd model, and pip install "langchain[mcp]" hands me every MCP server I already run as a tool list. Nothing in that path touches a login.

What I dislike is reading the code: the abstraction layers are deep and a stack trace crosses more packages than I would like. But I can read it, and I have patched it locally more than once. It goes where I point it, offline included. That is the bar.