Code Agents: The Future of Agentic AI

of AI brokers. LLMs are not simply instruments. They’ve grow to be energetic contributors in our lives, boosting productiveness and remodeling the way in which we reside and work. 

• OpenAI lately launched Operator, an AI agent that may autonomously carry out varied duties, from searching the net to filling out types and scheduling appointments.
• Anthropic launched MCP (Model Context Protocol), a brand new customary for the way AI assistants work together with the skin world. With over 5 thousand energetic MCP servers already, adoption is rising quickly.
• AI brokers are additionally altering the panorama of software program engineering. Instruments like GitHub Copilot’s agentic mode, Claude Code, OpenAI Codex, and others should not solely improving developer productivity and code quality but additionally democratising the sector, making software program growth accessible to folks with out a technical background.

We’ve beforehand checked out completely different AI Agent frameworks, comparable to LangGraph or CrewAI. On this article, I wish to focus on a brand new one I’ve been exploring lately — HuggingFace smolagents. It’s an attention-grabbing framework because it implements the idea of code brokers.

On this article, we’ll discover a number of subjects:

• What code brokers are (teaser: it’s not associated to vibe coding).
• Learn how to use the HuggingFace smolagents framework in apply. 
• Whether or not it’s safe to present LLMs a lot company.
• The true distinction in efficiency between code brokers and conventional tool-calling brokers.

AI Brokers recap

Let’s begin with a fast refresher: what precisely are AI brokers? HuggingFace provides a transparent and concise definition of what they imply by brokers. 

AI Brokers are applications the place LLM outputs management the workflow.

So, we’d like an agentic circulation after we desire a system to purpose and act based mostly on observations. Truly, company isn’t a binary variable (sure or no), however a spectrum. 

• At one finish, we are able to have techniques with out company in any respect, for instance, a easy course of the place an LLM defines the sentiment of a textual content, interprets it or summarises it. 
• The subsequent degree is routing, the place an LLM can classify an incoming query and resolve which path to take —  for instance, calling a device if a buyer is asking in regards to the standing of their present order, and transferring the dialog to a human CS agent in any other case. 
• Extra superior techniques can exhibit increased levels of company. These may embody the power to execute different LLMs (multi-agent setup) and even create new instruments on the fly.

Code brokers fall into this extra superior class. They’re multi-step brokers that execute device calls within the type of code, in distinction to the extra conventional method utilizing a JSON format with the device title and arguments.

A number of latest papers have proven that utilizing code in agentic flows results in higher outcomes: 

It is smart when you consider it. We’ve been creating programming languages for many years to resolve complicated issues. So, it’s pure that these languages are higher suited to LLM’s duties than easy JSON configs. An extra profit is that LLMs are already fairly good at writing code in widespread programming languages, because of the huge quantity of accessible knowledge for coaching.

This method comes with a number of different advantages as effectively:

• By producing code, an LLM isn’t restricted to a predefined set of instruments and may create its personal features.
• It may well mix a number of instruments inside a single motion utilizing circumstances and loops, which helps scale back the variety of steps required to finish a activity.  
• It additionally allows the mannequin to work with a greater diversity of outputs, comparable to producing charts, pictures, or different complicated objects.

These advantages aren’t simply theoretical; we are able to observe them in apply. In “Executable Code Actions Elicit Better LLM Agents”, the authors present that code brokers outperform conventional strategies, attaining the next success fee and finishing a activity in fewer steps, which in flip reduces prices. 

Code brokers look promising, which impressed me to do this method in apply.

HuggingFace smolagents framework

First strive

Fortunately, we don’t have to construct code brokers from scratch, as HuggingFace has launched a helpful library referred to as smolagents that implements this method. 

Let’s begin by putting in the library. 

pip set up smolagents[litellm] 
# I've used litellm, since I am planning to make use of it with OpenAI mannequin

Subsequent, let’s construct a fundamental instance. To initialise the agent, we’d like simply two parameters: mannequin and instruments. 

I plan to make use of OpenAI for the mannequin, which is accessible through LiteLLM. Nonetheless, the framework helps different choices as effectively. You need to use an area mannequin through Ollama or TransformersModel, or public fashions through Inference Providers or select different choices (yow will discover extra particulars in the documentation). 

I didn’t specify any instruments, however used add_base_tools = True, so my agent has a default set of tools, comparable to a Python interpreter or DuckDuckGo search. Let’s strive it out with a easy query.

from smolagents import CodeAgent, LiteLLMModel
mannequin = LiteLLMModel(model_id="openai/gpt-4o-mini", 
  api_key=config['OPENAI_API_KEY']) 
agent = CodeAgent(instruments=[], mannequin=mannequin, add_base_tools=True)

agent.run(
    """I've 5 completely different balls and I randomly choose 2. 
    What number of attainable combos of the balls I can get?""",
)

In consequence, we see a very properly formatted execution circulation. It’s simply superb and means that you can perceive the method completely.

So, the agent discovered a solution in a single step and wrote Python code to calculate the variety of combos. 

The output is kind of useful, however we are able to go even deeper and take a look at the entire info associated to execution (together with prompts), through agent.reminiscence.steps. Let’s take a look at the system immediate utilized by the agent. 

You might be an professional assistant who can remedy any activity utilizing code blobs. 
You may be given a activity to resolve as greatest you possibly can.

To take action, you've got been given entry to a listing of instruments: these instruments 
are principally Python features which you'll name with code.

To resolve the duty, you could plan ahead to proceed in a sequence of 
steps, in a cycle of 'Thought:', 'Code:', 
and 'Commentary:' sequences.

At every step, within the 'Thought:' sequence, it's best to first clarify 
your reasoning in the direction of fixing the duty and the instruments that you really want 
to make use of.

Then within the 'Code:' sequence, it's best to write the code in easy 
Python. The code sequence should finish with '<end_code>' sequence.

Throughout every intermediate step, you should utilize 'print()' to save lots of 
no matter necessary info you'll then want.

These print outputs will then seem within the 'Commentary:' area, 
which will probably be out there as enter for the subsequent step.

Ultimately you must return a last reply utilizing 
the final_answer device.

Listed below are a couple of examples utilizing notional instruments: <...>

It’s fairly clear that smolagents implements the ReAct method (launched within the paper by Yao et al. “ReAct: Synergizing Reasoning and Acting in Language Models”) and makes use of a few-shot prompting approach.

The smolagents library handles all behind-the-scenes work concerned within the agent workflow: assembling the system immediate with all essential info for the LLM (i.e. out there instruments), parsing the output and executing the generated code. It additionally supplies complete logging and a retry mechanism to assist right errors.

Moreover, the library presents reminiscence administration options. By default, all execution outcomes are saved to reminiscence, however you possibly can customise this behaviour. For instance, you possibly can take away some middleman outcomes from the reminiscence to scale back the variety of tokens or execute the agent step-by-step. Whereas we received’t dive deep into reminiscence administration right here, yow will discover helpful code examples in the documentation.

Safety

Now, it’s time to debate the drawbacks of the code brokers’ method. Giving an LLM extra company by permitting it to execute arbitrary code introduces increased dangers. Certainly, an LLM can run dangerous code both by mistake (since LLMs are nonetheless removed from excellent) or attributable to focused assaults like immediate injections or compromised fashions. 

To mitigate these dangers, the native Python executor carried out within the smolagents library has a bunch of security checks: 

• By default, imports should not allowed except the bundle has been explicitly added to additional_authorized_imports record. 
• Furthermore, submodules are blocked by default, so you could authorise them particularly (i.e. numpy.* ). It’s been achieved as a result of some packages can expose doubtlessly dangerous submodules, i.e. random._os . 
• The whole variety of executed operations is capped, stopping infinite loops and useful resource bloating. 
• Any operation not explicitly outlined within the interpreter will elevate an error. 

Let’s take a look at whether or not these security measures really work.

from smolagents.local_python_executor import LocalPythonExecutor

custom_executor = LocalPythonExecutor(["numpy.*", "random"])

# operate to have fairly formatted exceptions
def run_capture_exception(command: str):
    strive:
        custom_executor(harmful_command)
    besides Exception as e:
        print("ERROR:n", e)

# Unauthorised imports are blocked 
harmful_command="import os; exit_code = os.system('<bad_command>')"
run_capture_exception(harmful_command)
# ERROR: Code execution failed at line 'import os' attributable to: 
# InterpreterError: Import of os isn't allowed. Approved imports 
# are: ['datetime', 'itertools', 're', 'math', 'statistics', 'time', 'queue', 
# 'numpy.*', 'random', 'collections', 'unicodedata', 'stat']

# Submodules are additionally blocked except said particularly
harmful_command="from random import _os; exit_code = _os.system('<bad_command>')"
run_capture_exception(harmful_command)
# ERROR: Code execution failed at line 'exit_code = _os.system('<bad_command>')' 
# attributable to: InterpreterError: Forbidden entry to module: os

# The cap on the variety of iterations breaks inifinity loops 
harmful_command = '''
whereas True: 
    go
'''
run_capture_exception(harmful_command)
# ERROR: Code execution failed at line 'whereas True: go' attributable to: 
# InterpreterError: Most variety of 1000000 iterations in Whereas loop 
# exceeded

# Undefined operations do not work
harmful_command="!echo <bad_command>"
custom_executor(harmful_command)
# ERROR: Code parsing failed on line 1 attributable to: SyntaxError

It appears we’ve some security nets with code brokers. Nonetheless, regardless of these safeguards, dangers persist whenever you’re executing code regionally. For instance, an LLM can recursively create threads in your laptop or create too many information, resulting in useful resource bloating. A attainable resolution is to execute code in a sandboxed surroundings, comparable to utilizing Docker or options like E2B. I’m prepared to be adventurous and run my code regionally, however if you happen to favor a extra risk-averse method, you possibly can comply with the sandbox set-up steerage in the documentation.

Code agent vs conventional Software-Calling agent

It’s claimed that the code brokers carry out higher in comparison with the normal JSON-based method. Let’s put this to the take a look at. 
I’ll use the duty of metrics change evaluation that I described in my earlier article, “Making sense of KPI changes”. We’ll begin with a simple case: analysing a easy metric (income) cut up by one dimension (nation).

raw_df = pd.read_csv('absolute_metrics_example.csv', sep = 't')
df = raw_df.groupby('nation')[['revenue_before', 'revenue_after_scenario_2']].sum()
  .sort_values('revenue_before', ascending = False).rename(
    columns = {'revenue_after_scenario_2': 'after', 
      'revenue_before': 'earlier than'})

The smolagents library helps two courses, which we are able to use to check two approaches: 

• CodeAgent — an agent that acts by producing and executing code,
• ToolCallingAgent — a standard JSON-based agent. 

Our brokers will want some instruments, so let’s implement them. There are multiple options to create tools in smolagents: we are able to re-use LangChain instruments, obtain them from HuggingFace Hub or just create Python features. We’ll take essentially the most easy method by writing a few Python features and annotating them with @device. 

I’ll create two instruments: one to estimate the relative distinction between metrics, and one other to calculate the sum of a listing. Since LLM will probably be utilizing these instruments, offering detailed descriptions is essential. 

@device 
def calculate_metric_increase(earlier than: float, after: float) -> float: 
    """
    Calculate the proportion change of the metric between earlier than and after

    Args:
        earlier than: worth earlier than
        after: worth after
    """
    return (earlier than - after) * 100/ earlier than

@device 
def calculate_sum(values: record) -> float: 
    """
    Calculate the sum of record

    Args:
        values: record of numbers
    """
    return sum(values)

Teaser: I’ll later realise that I ought to have supplied extra instruments to the agent, however I genuinely neglected them.

CodeAgent

Let’s begin with a CodeAgent. I’ve initialised the agent with the instruments we outlined earlier and authorised the utilization of some Python packages that is likely to be useful. 

agent = CodeAgent(
    mannequin=mannequin,
    instruments=[calculate_metric_increase, calculate_sum],
    max_steps=10,
    additional_authorized_imports=["pandas", "numpy", "matplotlib.*", 
      "plotly.*"],
    verbosity_level=1 
)

activity = """
Here's a dataframe exhibiting income by phase, evaluating values 
earlier than and after.
Might you please assist me perceive the adjustments? Particularly:
1. Estimate how the entire income and the income for every phase 
have modified, each in absolute phrases and as a proportion.
2. Calculate the contribution of every phase to the entire 
change in income.

Please spherical all floating-point numbers within the output 
to 2 decimal locations.
"""

agent.run(
    activity,
    additional_args={"knowledge": df},
)

General, the code agent accomplished the duty in simply two steps, utilizing solely 5,451 enter and 669 output tokens. The end result additionally seems fairly believable.

{'total_before': 1731985.21, 'total_after': 
1599065.55, 'total_change': -132919.66, 'segment_changes': 
{'absolute_change': {'different': 4233.09, 'UK': -4376.25, 'France': 
-132847.57, 'Germany': -690.99, 'Italy': 979.15, 'Spain': 
-217.09}, 'percentage_change': {'different': 0.67, 'UK': -0.91, 
'France': -55.19, 'Germany': -0.43, 'Italy': 0.81, 'Spain': 
-0.23}, 'contribution_to_change': {'different': -3.18, 'UK': 3.29, 
'France': 99.95, 'Germany': 0.52, 'Italy': -0.74, 'Spain': 0.16}}}

Let’s check out the execution circulation. The LLM acquired the next immediate. 

╭─────────────────────────── New run ────────────────────────────╮
│                                                                │
│ Here's a pandas dataframe exhibiting income by phase,         │
│ evaluating values earlier than and after.                             │
│ Might you please assist me perceive the adjustments?               │
│ Particularly:                                                  │
│ 1. Estimate how the entire income and the income for every     │
│ phase have modified, each in absolute phrases and as a          │
│ proportion.                                                    │
│ 2. Calculate the contribution of every phase to the entire     │
│ change in income.                                             │
│                                                                │
│ Please spherical all floating-point numbers within the output to 2   │
│ decimal locations.                                                │
│                                                                │
│ You will have been supplied with these further arguments, that   │
│ you possibly can entry utilizing the keys as variables in your python      │
│ code:                                                          │
│ {'df':             earlier than      after                           │
│ nation                                                        │
│ different    632767.39  637000.48                                  │
│ UK       481409.27  477033.02                                  │
│ France   240704.63  107857.06                                  │
│ Germany  160469.75  159778.76                                  │
│ Italy    120352.31  121331.46                                  │
│ Spain     96281.86   96064.77}.                                │
│                                                                │
╰─ LiteLLMModel - openai/gpt-4o-mini ────────────────────────────╯

In step one, the LLM generated a dataframe and carried out all calculations. Apparently, it selected to write down all of the code independently fairly than utilizing the supplied instruments. 

Much more surprisingly, the LLM recreated the dataframe based mostly on the enter knowledge as an alternative of referencing it straight. This method isn’t perfect (particularly when working with large datasets), as it could possibly result in errors and better token utilization. This behaviour might doubtlessly be improved through the use of a extra specific system immediate. Right here’s the code the agent executed in step one.

import pandas as pd                                                                                                        
 
# Creating the DataFrame from the supplied knowledge                 
knowledge = {                                                        
    'earlier than': [632767.39, 481409.27, 240704.63, 160469.75,      
120352.31, 96281.86],                                           
    'after': [637000.48, 477033.02, 107857.06, 159778.76,       
121331.46, 96064.77]                                            
}                                                               
index = ['other', 'UK', 'France', 'Germany', 'Italy', 'Spain']  
df = pd.DataFrame(knowledge, index=index)                            
                                                                
# Calculating whole income earlier than and after                    
total_before = df['before'].sum()                               
total_after = df['after'].sum()                                 
                                                                
# Calculating absolute and proportion change for every phase   
df['absolute_change'] = df['after'] - df['before']              
df['percentage_change'] = (df['absolute_change'] /              
df['before']) * 100                                             
                                                                
# Calculating whole income change                              
total_change = total_after - total_before                       
                                                                
# Calculating contribution of every phase to the entire change  
df['contribution_to_change'] = (df['absolute_change'] /         
total_change) * 100                                             
                                                                
# Rounding outcomes                                              
df = df.spherical(2)                                                
                                                                
# Printing the calculated outcomes                               
print("Complete income earlier than:", total_before)                    
print("Complete income after:", total_after)                      
print("Complete change in income:", total_change)                 
print(df)

Within the second step, the LLM merely constructed the ultimate reply by referring to the variables calculated on the earlier step (which is admittedly neat). 

final_answer({                                                  
    "total_before": spherical(total_before, 2),                     
    "total_after": spherical(total_after, 2),                       
    "total_change": spherical(total_change, 2),                     
    "segment_changes": df[['absolute_change',                   
'percentage_change', 'contribution_to_change']].to_dict()       
})

It labored fairly effectively.

ToolCallingAgent

Now, it’s time to see how conventional tool-calling brokers can sort out this drawback. We initialised it in an analogous method and ran the duty.

from smolagents import ToolCallingAgent
traditional_agent = ToolCallingAgent(
    mannequin=mannequin,
    instruments=[calculate_metric_increase, calculate_sum],
    max_steps=30,
)

activity = """
Here's a dataframe exhibiting income by phase, evaluating values 
earlier than and after.
Might you please assist me perceive the adjustments? Particularly:
1. Estimate how the entire income and the income for every phase 
have modified, each in absolute phrases and as a proportion.
2. Calculate the contribution of every phase to the entire 
change in income.

Please spherical all floating-point numbers within the output 
to 2 decimal locations.
"""

traditional_agent.run(
    activity,
    additional_args={"knowledge": df},
)

The outcomes are removed from perfect: solely the relative adjustments are right, whereas the remainder of the numbers are pure hallucinations. I’ve to confess, the core subject was the shortage of acceptable instruments (particularly, instruments to calculate variations and to estimate shares). Nonetheless, the agent ought to have flagged lacking instruments fairly than producing random numbers.

Complete income change: -7319.66 (-7.67%). Income Adjustments by Section:
- Different: +232.09 (-0.67%)
- UK: -4376.25 (0.91%)
- France: -132847.57 (55.19%)
- Germany: -690.99 (0.43%)
- Italy: +979.15 (-0.81%)
- Spain: -217.09 (0.23%)
Contribution to whole change:
- Different: 0.03%
- UK: -59.88%
- France: -181.77%
- Germany: -9.43%
- Italy: +13.38%
- Spain: -0.03%

By way of useful resource utilization, the tool-calling agent carried out considerably worse: 12 steps, with 29,201 enter and 1,695 output tokens. So, code brokers clearly provide price financial savings in terms of agent execution.

Let’s dig a bit deeper to know what the agent really did. First, it took 4 iterations simply to determine methods to name calculate_sum accurately and estimate the entire income earlier than and after. 

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Step 1 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
╭──────────────────────────────────────────────────────────────╮
│ Calling device: 'calculate_sum' with arguments:                │
│ {'values': {'earlier than': [632767.39, 481409.27, 240704.63,      │
│ 160469.75, 120352.31, 96281.86],                             │
│ 'after': [637000.48, 477033.02, 107857.06, 159778.76,        │
│ 121331.46, 96064.77]}}                                       │
╰──────────────────────────────────────────────────────────────╯
Invalid name to device 'calculate_sum' with arguments           
{"values": {"earlier than": [632767.39, 481409.27, 240704.63,       
160469.75, 120352.31, 96281.86],                              
"after": [637000.48, 477033.02, 107857.06, 159778.76,         
121331.46, 96064.77]}}:                                       
unsupported operand kind(s) for +: 'int' and 'str'            
It's best to name this device with right enter arguments.       
Anticipated inputs: {"values": {"kind": "object",                
"description": "record of values"}}                             
Returns output kind: quantity                                   
Software description: 'Calculate the sum of record'                 
[Step 1: Duration 4.06 sec | Input tokens: 1,190 | Output: 250]

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Step 2 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
╭──────────────────────────────────────────────────────────────╮
│ Calling device: 'calculate_sum' with arguments:                │
│ {'values': {'earlier than': [632767.39, 481409.27, 240704.63,      │
│ 160469.75, 120352.31, 96281.86],                             │
│ 'after': [637000.48, 477033.02, 107857.06, 159778.76,        │
│ 121331.46, 96064.77]}}                                       │
╰──────────────────────────────────────────────────────────────╯
Invalid name to device 'calculate_sum' with arguments           
{"values": {"earlier than": [632767.39, 481409.27, 240704.63,       
160469.75, 120352.31, 96281.86],                              
"after": [637000.48, 477033.02, 107857.06, 159778.76,         
121331.46, 96064.77]}}:                                       
unsupported operand kind(s) for +: 'int' and 'str'            
It's best to name this device with right enter arguments.       
Anticipated inputs: {"values": {"kind": "object",                
"description": "record of values"}}                             
Returns output kind: quantity                                   
Software description: 'Calculate the sum of record'                 
[Step 2: Duration 8.60 sec | Input tokens: 2,815 | Output: 386]

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Step 3 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
╭──────────────────────────────────────────────────────────────╮
│ Calling device: 'calculate_sum' with arguments:                │
│ {'values': [632767.39, 481409.27, 240704.63, 160469.75,      │
│ 120352.31, 96281.86]}                                        │
╰──────────────────────────────────────────────────────────────╯
Observations: 1731985.2100000002                              
[Step 3: Duration 1.23 sec | Input tokens: 4,871 | Output: 488]

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Step 4 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
╭──────────────────────────────────────────────────────────────╮
│ Calling device: 'calculate_sum' with arguments:                │
│ {'values': [637000.48, 477033.02, 107857.06, 159778.76,      │
│ 121331.46, 96064.77]}                                        │
╰──────────────────────────────────────────────────────────────╯
Observations: 1599065.55                                                                                                                      

The subsequent seven steps have been spent calculating the relative metric adjustments utilizing the calculate_metric_increase device.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Step 5 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
╭──────────────────────────────────────────────────────────────╮
│ Calling device: 'calculate_metric_increase' with              │
│ arguments: {'earlier than': 1731985.21, 'after': 1599065.55}       │
╰──────────────────────────────────────────────────────────────╯
Observations: 7.674410799385517                               

<-- related device requires all nation segments -->            

Ultimately, the agent put collectively a last name. 

So, if the LLM had had instruments to calculate absolutely the distinction and the share of the sum, it will have taken an extra 14 iterations and much more tokens. After all, we are able to stop such inefficiencies by fastidiously designing the instruments we offer:

• We might modify our features to work with lists of values as an alternative of single gadgets, which might considerably scale back the variety of steps. 
• Moreover, we might create extra complicated features that calculate all essential metrics directly (much like what the code agent did). This fashion, LLM wouldn’t have to carry out calculations step-by-step. Nonetheless, this method may scale back the pliability of the system. 

Though the outcomes weren’t perfect attributable to a poor alternative of instruments, I nonetheless discover this instance fairly insightful. It’s clear that code brokers are extra highly effective, cost-efficient and versatile as they’ll invent their very own complete instruments and carry out a number of actions in a single step.

You could find the entire code and execution logs on GitHub.

Abstract

We’ve realized rather a lot in regards to the code brokers. Now, it’s time to wrap issues up with a fast abstract.

Code brokers are LLM brokers that “assume” and act utilizing Python code. As a substitute of calling instruments through JSON, they generate and execute precise code. It makes them extra versatile and cost-efficient as they’ll invent their very own complete instruments and carry out a number of actions in a single step. 

HuggingFace has introduced this lifestyle of their framework, smolagents. Smolagents makes it simple to construct fairly complicated brokers with out a lot problem, whereas additionally offering security measures throughout the code execution. 

On this article, we’ve explored the essential performance of the smolagents library. However there’s much more to it. Within the subsequent article, we’ll dive into extra superior options (like multi-agent setup and planning steps) to construct the agent that may narrate KPI adjustments. Keep tuned!

Thank you numerous for studying this text. I hope this text was insightful for you.

Reference

This text is impressed by the “Building Code Agents with Hugging Face smolagents” quick course by DeepLearning.AI.