Creating tools for agents in LangChain - Performance Optimization Steps

Performance: Creating tools for agents

MEDIUM IMPACT

This affects the responsiveness and efficiency of agent interactions by controlling how external tools are integrated and called during runtime.

Integrating external tools for agent queries

LangChain

import asyncio

class AsyncTool:
    async def run(self, input):
        await asyncio.sleep(0.1)  # Simulates fast async response
        return f"Processed {input}"

agent = Agent(tools=[AsyncTool()])
response = await agent.run('query')

Using asynchronous calls allows the agent to handle other tasks while waiting, reducing blocking time.

📈 Performance GainReduces blocking from 2 seconds to 0.1 seconds, improving INP and user experience

Integrating external tools for agent queries

LangChain

class SlowTool:
    def run(self, input):
        import time
        time.sleep(2)  # Simulates slow response
        return f"Processed {input}"

agent = Agent(tools=[SlowTool()])
response = agent.run('query')

The tool blocks the agent for 2 seconds per call, causing slow responses and poor user experience.

📉 Performance CostBlocks interaction for 2 seconds per tool call, increasing INP significantly

Performance Comparison

Pattern	DOM Operations	Reflows	Paint Cost	Verdict
Synchronous blocking tool call	Minimal	0	Blocks interaction for seconds	[X] Bad
Asynchronous non-blocking tool call	Minimal	0	Minimal delay, smooth interaction	[OK] Good

Rendering Pipeline

When an agent calls a tool, the browser or runtime waits for the tool's response before continuing. Slow or blocking tools delay the interaction to next paint (INP). Async tools allow the runtime to continue processing other tasks, improving responsiveness.

→Interaction Handling

→JavaScript Execution

→Rendering

⚠️ BottleneckJavaScript Execution waiting on synchronous tool calls

Core Web Vital Affected

INP

This affects the responsiveness and efficiency of agent interactions by controlling how external tools are integrated and called during runtime.

Optimization Tips

1Avoid synchronous blocking calls in agent tools to keep interactions responsive.

2Use asynchronous programming to allow parallel processing during tool calls.

3Lazy load tools only when needed to reduce initial load and improve responsiveness.

Performance Quiz - 3 Questions

What is the main performance issue with synchronous tool calls in agents?

A They cause excessive CSS recalculations.

B They block the main thread, causing slow interaction response.

C They increase the number of DOM nodes unnecessarily.

D They reduce bundle size too much.

DevTools: Performance

How to check: Record a performance profile while triggering the agent tool call. Look for long tasks or scripting blocking the main thread.

What to look for: Long scripting tasks or gaps in responsiveness indicate blocking synchronous calls; short tasks with idle time indicate async efficiency.

Practice

1. What is the main purpose of creating tools for agents in Langchain?

easy

A. To let agents perform specific tasks easily

B. To make the code run faster

C. To store data permanently

D. To create user interfaces

Solution

1. Step 1: Understand the role of tools in Langchain

   Tools are designed to help agents do tasks by providing specific functions.

2. Step 2: Identify the main benefit

   By using tools, agents can perform tasks more easily and effectively.

3. Final Answer:

   To let agents perform specific tasks easily -> Option A

4. Quick Check:

   Tools help agents do tasks = B [OK]

Hint: Tools help agents do tasks simply and clearly [OK]

Common Mistakes:

• Thinking tools speed up code execution
• Confusing tools with data storage
• Assuming tools create user interfaces

2. Which of the following is the correct way to create a tool in Langchain?

easy

A. tool = Tool(func=search_function)

B. tool = Tool('search', description='Searches data')

C. tool = Tool(name='search', description='Searches data')

D. tool = Tool(name='search', func=search_function, description='Searches data')

Solution

1. Step 1: Check required parameters for Tool

   The Tool constructor needs a name, a function (func), and a description.

2. Step 2: Match parameters with options

   Only tool = Tool(name='search', func=search_function, description='Searches data') provides all three: name, func, and description correctly.

3. Final Answer:

   tool = Tool(name='search', func=search_function, description='Searches data') -> Option D

4. Quick Check:

   Tool needs name, func, description = C [OK]

Hint: Tool needs name, function, and description to work [OK]

Common Mistakes:

• Omitting the function parameter
• Passing parameters in wrong order
• Leaving out the description

3. Given this code snippet, what will be the output when the agent uses the tool?

def greet(name: str) -> str:
    return f"Hello, {name}!"

from langchain.agents import Tool

greet_tool = Tool(name='greet', func=greet, description='Greets a person')

result = greet_tool.func('Alice')
print(result)

medium

A. "greet Alice"

B. "Hello, Alice!"

C. Error: Missing agent call

D. "Hello!"

Solution

1. Step 1: Understand the greet function behavior

   The greet function returns a string "Hello, {name}!" with the given name.

2. Step 2: Check how the tool is used

   The tool calls greet with 'Alice', so it returns "Hello, Alice!" which is printed.

3. Final Answer:

   "Hello, Alice!" -> Option B

4. Quick Check:

   greet('Alice') = "Hello, Alice!" [OK]

Hint: Tool calls function directly with given input [OK]

Common Mistakes:

• Thinking the tool prints 'greet Alice'
• Assuming an error without agent context
• Ignoring the name parameter in output

4. Identify the error in this tool creation code:

def add_numbers(a, b):
    return a + b

from langchain.agents import Tool

add_tool = Tool(name='add', func=add_numbers, description='Adds two numbers')

result = add_tool.func(5)
print(result)

medium

A. Missing one argument when calling add_tool.func

B. Tool name must be uppercase

C. Description is too short

D. Function add_numbers should not return a value

Solution

1. Step 1: Check function parameters

   add_numbers requires two arguments: a and b.

2. Step 2: Check function call

   add_tool.func is called with only one argument (5), missing the second argument.

3. Final Answer:

   Missing one argument when calling add_tool.func -> Option A

4. Quick Check:

   Function needs 2 args, call has 1 = D [OK]

Hint: Match function parameters with call arguments count [OK]

Common Mistakes:

• Ignoring argument count mismatch
• Thinking tool name case matters
• Assuming description length causes error

5. You want to create a tool that converts temperatures from Celsius to Fahrenheit for an agent. Which code correctly creates this tool with a clear description?

hard

A. def convert_temp(f):
return (f - 32) * 5/9

from langchain.agents import Tool

tool = Tool(name='temp_convert', func=convert_temp, description='Converts Fahrenheit to Celsius')

B. def c_to_f(c):
return c + 32

from langchain.agents import Tool

temp_tool = Tool(name='temp', func=c_to_f, description='Temperature conversion')

C. def c_to_f(c):
return (c * 9/5) + 32

from langchain.agents import Tool

temp_tool = Tool(name='temp_convert', func=c_to_f, description='Converts Celsius to Fahrenheit')

D. def c_to_f(c):
return (c * 9/5) + 32

tool = Tool(func=c_to_f, description='Convert temp')

Solution

1. Step 1: Verify function correctness

   The c_to_f(c) function returns (c * 9/5) + 32, which correctly converts Celsius to Fahrenheit using the standard formula.

2. Step 2: Check tool creation parameters

   Tool(name='temp_convert', func=c_to_f, description='Converts Celsius to Fahrenheit') uses name, func, and a clear description matching the task.

3. Final Answer:

   def c_to_f(c):\n return (c * 9/5) + 32\n\nfrom langchain.agents import Tool\n\ntemp_tool = Tool(name='temp_convert', func=c_to_f, description='Converts Celsius to Fahrenheit') -> Option C

4. Quick Check:

   Correct formula and clear tool setup = A [OK]

Hint: Use correct formula and full tool parameters [OK]

Common Mistakes:

• Using wrong temperature formula
• Omitting tool name or description
• Confusing Celsius to Fahrenheit with reverse