LangChain Agents 1 · From a bare model call to an agent with hands

[akash-coded]

LangChain Agents · Exercise 1

Language: Python Topics: the agent loop, message types, bare vs tool agents, reading LangChain and Strands syntax Level: Foundational

Answers are letters, pairs like 1-C, T/F, or a sequence. You read code here, you do not write any.

---

Q1 · (2 pts) A language model, stripped to its core, is:

• A) a fact database it queries
• B) a next-token predictor that samples from a probability distribution
• C) a keyword-to-intent rules engine
• D) a search index over training data

---

Q2 · Trace the loop (3 pts) Edges are tagged a b c d. Start at the user message.

flowchart TD
    U[user message] -->|a| M[model]
    M -->|b tool needed| T[run tool]
    T -->|c| M
    M -->|d no tool| F[final answer]

Loading

Right after the tool runs, which edge fires?

• A) a B) b C) c D) d

---

Q3 · One arrow is flipped (3 pts)

flowchart TD
    U[user message] -->|a| M[model]
    M -->|b tool needed| T[run tool]
    T -->|c| U
    M -->|d no tool| F[final answer]

Loading

Which tagged edge is wrong?

• A) a B) b C) c D) d

---

Q4 · Read the code (3 pts)

model = ScriptedChatModel(responses=[AIMessage(content="How can I help?")])
agent = create_agent(model, tools=[], system_prompt="You are TravelMind.")
result = agent.invoke({"messages": [{"role": "user", "content": "Hi"}]})

How many messages are in result["messages"]?

• A) 1 B) 2 C) 3 D) 4

---

Q5 · Read the code (3 pts)

AIMessage(content="", tool_calls=[{"name": "lookup_pnr", "args": {"pnr": "JX48Q2"}, "id": "t1", "type": "tool_call"}])

What is the model doing here?

• A) returning the final answer
• B) asking to run lookup_pnr with pnr="JX48Q2"
• C) raising an error
• D) nothing, the content is empty

---

Q6 · Match the message to its job (4 pts)

Message type		Its job
1. SystemMessage		A. the user's request
2. HumanMessage		B. standing instructions, set once
3. AIMessage		C. an answer, or a request to call a tool
4. ToolMessage		D. the result of a tool the model asked for
		E. a raw string with no role

---

Q7 · Pick the correct bare agent (2 pts) A bare agent has no tools.

Option A

flowchart LR
    U[user] --> M[model] --> A[answer]

Loading

Option B

flowchart LR
    U[user] --> M[model] --> T[tools node] --> A[answer]

Loading

---

Q8 · Order the turn (3 pts) One tool-using turn runs these. Put them in order.

• a) the tools node runs the requested tool
• b) the user message enters the loop
• c) the model reads the tool result and writes the answer
• d) the model requests a tool by name and arguments

---

Q9 · Spot the line that does not belong (3 pts) In LangChain 1.0:

L1  from langchain.agents import create_agent
L2  from langchain.agents import AgentExecutor
L3  from langchain.tools import tool
L4  from langgraph.checkpoint.memory import InMemorySaver

• A) L1 B) L2 C) L3 D) L4

---

Q10 · Same idea, two frameworks (4 pts) You know the left from Strands. Match each to its LangChain twin.

Strands		LangChain
1. agent("...")		A. ChatBedrockConverse(model=..., region_name=...)
2. str(result)		B. agent.invoke({"messages": [...]})
3. BedrockModel(model_id=..., region_name=...)		C. result["messages"][-1].content
		D. create_agent(model, tools, system_prompt=)

---

Q11 · Pick all that apply (3 pts) Which are real LangChain message types?

• A) SystemMessage
• B) HumanMessage
• C) PromptMessage
• D) AIMessage
• E) ToolMessage
• F) UserBlock

---

Q12 · True or false (3 pts) Mark each T or F.

1. A bare agent still reaches the tools node on every run.
2. The tools node runs the code, the model only asks for it.
3. A transcript is a growing list of typed messages.

---

Q13 · Read the model output (2 pts) The us. prefix on us.anthropic.claude-haiku-4-5-20251001-v1:0 is there because:

• A) it names a US-only model
• B) Bedrock requires the inference-profile prefix to route the call
• C) it is optional styling
• D) it sets the region

---

Case study · TravelMind day one (5 pts)

TravelMind launches with a bare agent, no tools. Rao asks: "is JX48Q2 cancelled?"

flowchart LR
    U[is JX48Q2 cancelled?] --> M[model, no tools] --> A[answer]

Loading

Q14a (2 pts) With no tool to check the booking, the agent is forced to:

• A) look the record up anyway
• B) guess from training data, ungrounded
• C) refuse to answer
• D) call a tool it does not have

Q14b (3 pts) The single change that grounds the answer:

• A) a longer system prompt
• B) a lookup_pnr tool wired into the agent
• C) higher temperature
• D) a larger model

[akash-coded]

from typing import Any, List
from langchain_core.language_models.chat_models import BaseChatModel
from langchain_core.messages import AIMessage
from langchain_core.outputs import ChatResult, ChatGeneration
from langchain.agents import create_agent
from langchain.tools import tool


class ScriptedChatModel(BaseChatModel):
    '''Deterministic stand-in. Real framework, scripted replies, no credentials.'''
    responses: List[Any]
    idx: int = 0

    @property
    def _llm_type(self) -> str:
        return "scripted"

    def _generate(self, messages, stop=None, run_manager=None, **kwargs):
        reply = self.responses[min(self.idx, len(self.responses) - 1)]
        object.__setattr__(self, "idx", self.idx + 1)
        return ChatResult(generations=[ChatGeneration(message=reply)])

    def bind_tools(self, tools, **kwargs):
        return self


@tool
def lookup_pnr(pnr: str) -> str:
    '''Return the booking status for a PNR.'''
    return {"JX48Q2": "BLR-DEL cancelled, Rao, Gold tier"}.get(pnr, "PNR not found")


# Bare agent: no tools, the loop turns once.
bare = create_agent(
    ScriptedChatModel(responses=[AIMessage(content="How can I help?")]),
    tools=[],
    system_prompt="You are TravelMind.",
)
print("bare:", bare.invoke({"messages": [{"role": "user", "content": "Hi"}]})["messages"][-1].content)

# Tool agent: the model asks for lookup_pnr, the tool answers, the model replies.
tool_agent = create_agent(
    ScriptedChatModel(responses=[
        AIMessage(content="", tool_calls=[{"name": "lookup_pnr", "args": {"pnr": "JX48Q2"}, "id": "t", "type": "tool_call"}]),
        AIMessage(content="JX48Q2 is cancelled. Gold tier, free rebooking."),
    ]),
    tools=[lookup_pnr],
    system_prompt="You are TravelMind.",
)
result = tool_agent.invoke({"messages": [{"role": "user", "content": "is JX48Q2 cancelled?"}]})
print("messages:", len(result["messages"]))     # 4: human, AI(tool call), tool, AI(answer)
print("final:", result["messages"][-1].content)

# Production swap:
# from langchain_aws import ChatBedrockConverse
# model = ChatBedrockConverse(model="us.anthropic.claude-haiku-4-5-20251001-v1:0", region_name="us-east-1")