The course, the exam, and the setup
Twenty-five percent of the exam, ten questions out of forty, drawn from two named courses. This is the larger one.
What you are actually being tested on
The official course is called Introduction to LangChain: Build AI Agents with Python, and the exam guide labels it “LangChain Foundations”. Thirty lessons, eighteen of them teaching, three modules with a build-along project closing each one.
The shape is deliberate and worth reading as a syllabus rather than a table of contents. Module 1 builds an agent that works. Module 2 makes it reach outside its own process. Module 3 makes it survivable in production. Every question in the Build section is somewhere on that arc, and knowing which of the three altitudes a question is asking about narrows the answer before you have read the options.
The mismatch you need to plan around
The official course is Python only. You write TypeScript for a living.
This is not a knowledge gap and it does not need extra study time. It is a dialect gap, and it is cheap to close deliberately and expensive to ignore. The concepts are identical across both languages: same hooks, same ordering rule, same thresholds, same jump targets. What differs is the spelling.
The setup, kept small
You need langchain and one provider package, and that is genuinely it for everything in this course. langchain-openai, langchain-anthropic, or langchain-google-genai, whichever key you already have. Add langgraph-checkpoint-postgres only when you get to persistence and want something that outlives the process.
Nothing in these sixteen lessons requires a deployment, a LangSmith organisation, or a cloud account.
The open-book part
The exam is semi-open-book, with docs.langchain.com and smith.langchain.com whitelisted.
Read that carefully, because it changes what is worth memorising. Finding a fact fast is a scoreable exam skill. Memorising a constructor signature is nearly worthless, because you can look it up in thirty seconds. What you cannot look up in thirty seconds is which of four plausible middlewares fits a described failure, because that requires you to already understand the catalogue.
So the study split that actually pays: memorise the distinctions and the numbers that are hard to search for, and practise navigating for everything else.
Try it yourself
What Build is worth
The exam guide gives the Build section a percentage and a question count. Both numbers are worth holding, because they tell you how much of your revision time this material deserves.
Show answer
Correct answer: B — 25 percent, 10 questions
Build is 25 percent of the exam and 10 of the 40 questions. The tempting wrong answer is the first one, because Build feels like the heart of the syllabus and people inflate it accordingly. It is a quarter, which is a lot but not a majority, and the other three sections between them are worth three times as much.
Two sources, one domain
The exam guide names more than one prep course under Build. Knowing that changes how you read a gap in this one.
How many official courses does the exam guide name for the Build section, and what does the other one cover?
Reveal answer
Two. This one, Introduction to LangChain, covers the create_agent harness: models, tools, memory, multimodal input, MCP, state and context, multi-agent, and middleware. The sibling course covers Deep Agents. So if a Build topic feels absent from this material, check whether it belongs to the Deep Agents half before assuming you have missed it.
The language the official course is written in
This is the structural fact about the course that costs marks if you ignore it.
Show answer
Correct answer: C — It is Python only, and the API names differ in case and shape from TypeScript
The official course is Python only. That matters because the naming conventions genuinely differ: Python uses snake_case parameters and PascalCase middleware classes, TypeScript uses camelCase for both. A TypeScript developer who learns the concepts but not the Python spelling will read a correct option as wrong on sight. The first answer is the comfortable assumption and it is not true.
Set up your open book before you need it
The exam is semi-open-book with two domains whitelisted, so navigation speed is a scoreable skill rather than a fallback. Build the bookmarks now, while you are calm.
- Open docs.langchain.com and bookmark the Python reference for agents, middleware, and tools as three separate bookmarks, not one folder.
- Add reference.langchain.com, which carries the full constructor signatures that the guide pages summarise and sometimes omit.
- Time yourself finding the default value of one middleware constructor parameter from a cold start. Anything over ninety seconds means your bookmarks are wrong, not that you are slow.
Where a constructor default actually lives
You need the exact default value of a middleware parameter and you have two minutes. Which surface answers it fastest?
Show answer
Correct answer: B — The reference page under reference.langchain.com, which lists signatures and defaults
The reference site carries constructor signatures with defaults; the guide pages explain purpose and frequently summarise parameters without their default values. The trace view is the tempting one, because a runtime value looks like ground truth, but it shows the value in effect for that run, which may have been overridden by config you did not write.