#23
Call limits for LangChain agents, explained easily
ModelCallLimitMiddleware caps model calls per run and per thread, and ToolCallLimitMiddleware caps each tool, so a looping agent stops before it drains the bill.
Episodes
Watch the series.
All topics
- #22
Cost per user in LangSmith, explained easily
Tag each run with a user_id in the config metadata and group the LangSmith dashboard by it to split cost, tokens and latency per customer. Check one trace by hand before trusting the split.
- #21
Guardrails for AI agents in LangChain, explained easily
Guardrails check model inputs, outputs and tool calls before they leave the system. Use fast fixed rules for clear patterns and model-based checks when meaning matters.
- #20
Evaluations in LangSmith, explained easily
A prompt change that feels better proves nothing. Run both versions against the same dataset, score them consistently, and compare improvements with regressions.
- #19
Checkpointing in LangGraph, explained easily
LangGraph checkpoints let crashed AI agents resume from their last successful step. External side effects still need retry-safe task design.
- #18
Context engineering in LangChain, explained easily
A capable model can still fail when it receives the wrong messages, tools or documents. Context engineering chooses the right information for the current step.
- #17
Long-term memory in LangChain, explained easily
Long-term memory carries saved facts across chats. Scope it incorrectly, and one user can retrieve another person's stored context.
- #16
Human in the loop in LangChain, explained easily
A capable agent acts, but a safe agent knows when to stop. Human-in-the-loop middleware pauses selected tool calls so a person can approve, edit or reject them before the agent resumes from a saved checkpoint.
- #15
Built-in middleware in LangChain, explained easily
You do not have to write middleware yourself. LangChain ships summarization, human in the loop, model fallback and a hard cap on model calls, ready to drop into the agent.
- #14
Middleware in LangChain, explained easily
The agent loop is fixed and you do not get to rewrite it. Middleware is your own code attached at the points it already has, before the model, after it answers, and around every tool call.
- #13
Structured output in LangChain, explained easily
Your agent answers in English, but your code needs a number, a date, a category. Hand it a schema and you get back a real validated object instead.
- #12
Streaming in LangChain, explained easily
An agent run can take 20 seconds. Streaming fills them, and there are three things you can stream: text, updates and custom.
- #11
What is event streaming? Clearly explained
A run emits text, tool calls and state all at once. Event streaming hands them over already separated, one typed channel each.
- #10
What is short term memory? Clearly explained
The thread id is the whole trick: reuse it and the conversation carries on, change it and the agent starts blank.
- #9
What are tools in LangChain? Clearly explained
The docstring is not a comment for your teammate, it is the only thing the model reads before deciding to call it.
- #8
What are messages in LangChain? Clearly explained
Everything the agent knows is one list, and it rereads the whole thing every turn.
- #7
Models in LangChain, explained easily
The one piece you did not build and cannot control, so what matters is how cheaply you can swap it.
- #6
What is an AI agent? Clearly explained
What separates an agent from a chatbot: it takes the next step, and knows when to stop and ask.
- #5
What is LangSmith? Clearly explained
How to see inside an agent once it is running, so you can trust what it did rather than hope.
- #4
What is LangGraph? Clearly explained
The runtime under the framework: state, checkpoints, and durable execution, in forty seconds.
- #3
What is LangChain? Clearly explained
The model is the engine and LangChain is the rest of the car: the wiring that turns a model into something that ships.
- #2
Is LangChain dead in 2026?
It still does 76 million downloads a week, so the interesting question is what changed rather than whether it died.
- #1
LangChain vs LangGraph vs Deep Agents: the 2026 map
LangChain renamed half its products in 2026, so here is what each piece is now called and what it actually does.
No matches
All series