LOCKEDIN / FORWARD-DEPLOYED ENGINEERING 101
Understand it.
Then build it.
You can be an experienced engineer and still be new to agents. Start with what each thing is and how it helps a person do useful work. Explore 58 original introductions across software, agents, cloud, security and delivery. Start here without a model key; each topic connects to deeper instruction and practice.
A place to start.
Choose a short reading route, or look up a concept below.
- 101UnderstandPlain language, visual walkthrough, vocabulary
- 102Build with guidanceOriginal lessons, examples and knowledge checks
- 103Practise and explainRunnable labs, failure cases and handover
Understand agents
Make the steps and shared state explicit when work needs to branch, wait or resume.
See the 8-concept route
For engineers who know software but are new to AI applications. Open the related definitions whenever a term is unfamiliar.
- 01The modelStart with what produces the text. Separate the model from the application using it.
- 02TokensSee the units a model reads and produces before reasoning about its input limits.
- 03ContextDecide what information reaches the model for this particular request.
- 04The agent loopPut the model inside an application that can execute permitted actions and check the result.
- 05WorkflowsSeparate a fixed sequence from a model-selected next action before choosing a framework.
- 06LangChainSee how a framework assembles the model, tools and application-controlled context.
- 07LangGraphYou are here
- 08MCPConnect capabilities through a shared protocol while keeping permission checks in the application and server.
Bookmark this page to return to this concept. The number shows your place in the reading route; it is not an assessment result.
An open-source orchestration framework and runtime
What is LangGraph?
LangGraph is a software framework for coordinating the steps of an application. You describe steps as a graph: what each step does, what information it carries forward and which step happens next. It is especially useful when work branches, waits for a person or needs to resume after a failure. You write code in Python or JavaScript; LangGraph is not a separate language or an AI model.
Client workflows rarely finish with one model call. An FDE may need to pause a draft for approval, keep the work while someone reviews it, and resume the same request without starting over.
UNDERSTAND THE SYSTEM
LangGraph runs a workflow around shared state
LangGraph is a software framework, used from Python or JavaScript. You write steps called nodes and routes called edges. The framework coordinates those steps around shared working data.
Read the connections in plain English
- 01Incoming request → Draft step
The workflow receives state such as request_id and the source request.
- 02Draft step → Human review
The draft and its version become the review target.
- 03Human review → Saved state
With persistence configured, the current state can survive the wait.
- 04Saved state → Route the decision
Resuming supplies the decision; application code checks that it matches the current proposal.
- 05Route the decision → Accepted or stopped
An accepted current version can continue. Rejected or stale decisions must not authorize execution.
Check your understanding: If the proposal changes to v4 while the reviewer approves v3, should the graph continue?
It should not execute v4 on a v3 approval. The application must detect the mismatch and request a new review. A workflow framework does not establish that business rule for you.
CONNECT IT TO WHAT YOU KNOW
Start with something familiar.
If you know a state machine or a build pipeline, start there. LangGraph lets you write named steps and transitions around shared working data, including work that waits for a person or survives a restart.
A graph is an explicit workflow representation. LangGraph is the framework that runs it. Python or JavaScript is the language you use to write it; the model is a separate component you may call from a step.
Words worth knowing
- Graph
- A map of steps and the connections between them.
- Node
- A function that performs one step.
- Edge
- A connection or rule that chooses the next step.
- State
- The information passed between steps, such as the request and current draft.
- Checkpoint
- A saved snapshot used to continue the work later.
- Interrupt
- A deliberate pause, often to wait for a person's input.
Useful foundations
Open a definition and diagram here. Your lesson stays in place.
Control flow with model decisionsWhat is an agentic workflow?
QUICK REFERENCE / 101
What is an agentic workflow?
A workflow is a defined arrangement of steps and decisions. An agentic workflow includes a model choosing some actions or paths, within application limits. A fixed workflow may use a model for one step without letting the model control the whole sequence.
An FDE must explain where the system can vary its behavior and where it must follow a rule.
An application built around an AI modelWhat is an AI agent?
QUICK REFERENCE / 101
What is an AI agent?
An AI agent is software that uses an AI model to choose some of the steps needed to complete a task. It can call tools, inspect their results and decide what to do next within limits set by the application. The model is one component; the agent also needs code, tools and rules about when to stop.
An FDE needs to decide when model-driven choices help a real task and when a predictable sequence is enough. An agent is useful only if it improves the work and stays within its allowed scope.
ONE EXAMPLE / FICTIONAL CASE
Follow the connections.
A teammate asks where to post a deployment update. Your program looks up a guide and prepares a suggested reply, or reports that the guide is missing. You could write this with functions and an if statement. LangGraph gives the functions names, carries their work record and follows the route you define. The review diagram shows how a later version can add a person to that flow.
UNDERSTAND THE SYSTEM
LangGraph runs a workflow around shared state
LangGraph is a software framework, used from Python or JavaScript. You write steps called nodes and routes called edges. The framework coordinates those steps around shared working data.
See what moves and why.
Read the connections in plain English
- 01Incoming request → Draft step
The workflow receives state such as request_id and the source request.
- 02Draft step → Human review
The draft and its version become the review target.
- 03Human review → Saved state
With persistence configured, the current state can survive the wait.
- 04Saved state → Route the decision
Resuming supplies the decision; application code checks that it matches the current proposal.
- 05Route the decision → Accepted or stopped
An accepted current version can continue. Rejected or stale decisions must not authorize execution.
Check your understanding: If the proposal changes to v4 while the reviewer approves v3, should the graph continue?
It should not execute v4 on a v3 approval. The application must detect the mismatch and request a new review. A workflow framework does not establish that business rule for you.
Work through it, step by step.
Start with the work record
Put the question and source identifier into a dictionary. The source and draft start empty. This is application state, not a model's memory.
What you have now: A request with source_id P1 and status new
Run a function and keep its updates
The lookup node is ordinary Python. It returns source and status; the graph keeps the other fields. The original question does not disappear.
What you have now: The source text and ready, or an empty source and needs_source
Choose the next step
The route reads the updated source. If present, draft formats a suggested reply. Otherwise, the graph ends. No model, review, sending or durable storage is needed for this first exercise.
What you have now: A draft or a clear missing-source result; neither means approved
A workflow that waits for a person
Original silent explainer · Optional English captions
Read the visual transcript
- Draft: a step writes a proposal and its version into shared state.
- Pause: save the state while a person reviews that exact proposal. Waiting is an explicit state, not a running model call.
- Resume: check that the decision matches the current proposal. Continue on acceptance; otherwise stop or request a new review.
A request becomes a draft, one field at a time
A teammate asks where to post a deployment update. One Python function looks up the guide; another prepares a suggested reply. LangGraph coordinates those functions and the record they share. Predict the result, then follow what each step changes.
- Node
- A function that does work
- State
- The work record it receives
- Edge
- The next step to run
The synthetic release guide exists. lookup finds it and draft prepares a suggested reply.
Source found → draft → END. Nothing sends the reply.
Put the request in a work record
Your calling code supplies five fields. The source and draft are empty because neither function has run. This dictionary is application state, not a prompt sent to a model.
See the Python for this step
graph.invoke({
"request": "Where do I post a deployment update?",
"source_id": "P1",
"source": "", "draft": "", "status": "new"
})Shared state after this step
Highlighted fields changed. Other fields stay in the work record.
- request changed
- Where do I post a deployment update?
- source_id changed
- P1
- source changed
- empty string
- draft changed
- empty string
- status changed
- new
Replay of the supplied lab’s reference runs with LangGraph 1.2.16. These controls inspect recorded state; run the Python file to execute the graph yourself. The draft is a fixed text template, with no model call or message sent.
FROM RECOGNITION TO REASONING
Explain two functions before adding a pause.
- 01
Choose P1, then P404. Predict whether a draft appears, the final status and whether the original request stays in the record.
- 02
Inspect lookup: it returns two fields. Then inspect route: it selects the next step without changing a field.
- 03
Run the linked Python lab. Add a synthetic P2 source in a separate copy and check that the draft uses its text.
Two predictions compared with the recorded runs, then your own local output and a tested source change.
Compare your reasoning when you are ready
A node is a function, state is the work record, and an edge determines what runs next. This graph keeps fields a node does not update. Both paths reach END, but only one prepares a draft. No model, review, sending or persistent storage is configured. Add those responsibilities deliberately in later lessons.
Keep notes in your own working copy. Moving through these cards does not record completion.
TURN THE IDEA INTO USEFUL WORK
You can now go deeper.
LangChain helps assemble model-and-tool agents. LangGraph gives you explicit control over the path and saved work. You can use LangGraph without LangChain, and graph steps do not all need AI.
Academy lessons use your existing training access. Free guides stay open.
Learn with the original creators
Start with Motivation and Simple Graph. Then compare its state-and-transition model with the illustration here. Provider registration may be required.