AgentBuilder vs n8n, the match of AI automation platforms

AgentBuilder offers real ease of use. On the other hand, it still remains limited for complex workflows compared to n8n.

Appeared at the beginning of October, AgentBuilder, which is part of OpenAI’s AgentKit, focuses on the creation of agentic workflows, like n8n. To get an idea of ​​the capacity of the two platforms, we compared them in terms of ease of use, wealth of options and prices. To do this, we designed two workflows with the same goal: being able to read emails using AI agents. It appears that AgentBuilder seems very interesting for native connectors, which were still few in number during our tests. It offers real ease of handling. On the other hand, it still remains limited for complex workflows. Its lack of customization possibilities penalizes it.

AgentBuilder promotes simplicity

We first set up an agent workflow on AgentBuilder. It allows you to read emails via chat. Basically, we connected a “Start” node to an “Agent” module. After choosing “Gmail” as the tool, we provided instructions for searching emails and clearly presenting the results. Our prompt:

Tu es un assistant Gmail intelligent. Tu as accès à l'outil "search_emails" pour chercher dans Gmail. Quand l'utilisateur demande de chercher des emails :
  1. Convertis sa demande en requête Gmail
  2. Appelle search_emails avec UNIQUEMENT query et max_results
  3. N'envoie PAS les champs tag et next_page_token
  4. Présente les résultats clairement
  SYNTAXE GMAIL :
  - from:nom (expéditeur)
  - after:YYYY/MM/DD (date)
  - newer_than:Xd (X jours)
  - is:unread (non lu)
  - in:inbox (boîte de réception)
  EXEMPLES D'APPEL :
  "dernier mail" → search_emails(query="in:inbox", max_results=1) "emails de bruno" → search_emails(query="from:bruno", max_results=10) "emails d'aujourd'hui" → search_emails(query="after:2024/10/13", max_results=10)
  IMPORTANT :
  - N'utilise QUE les paramètres query et max_results
  - Ne mets PAS tag ou next_page_token
  PRÉSENTATION DES RÉSULTATS :
  **Email X**
  • Expéditeur : (nom) <(email)>
  • Sujet : (sujet)
  • Date : (date)
  • Extrait : (extrait) Sois concis et professionnel.

The output format is “text”.

We see that it works properly.

In terms of handling, according to our tests, AgentBuilder allows you to create workflows that are sometimes simpler than those of n8n. No need for a Transform node to prepare data, a Code node to filter, or a separate OpenAI node. Thanks to “Function Calling”, everything is done by a single intelligent Agent. He knows when and how to use the Gmail tool. It automatically generates parameters and automatically formats results

On the other hand, developing a workflow requires advanced authentication with n8n, in order to operate Preview mode. Additionally, based on our testing, AgentBuilder’s non-native connectors are particularly difficult to integrate, compared to those of n8n. Note that the help is also less present than on n8n.

In terms of the wealth of options, AgentBuilder is generally below n8n. If it is possible in our workflow to read, search, filter emails, organize (label, archiving, deletion), or filter in an advanced way via natural language (temporal, sender, subject), the connectors are fewer than on n8n. There is no ability to write custom code in JavaScript or Python. Multiple integrations, whether to Slack, Notion or Airtable were also not present.

Remember that using the visual interface to design and test agents is free. Billing begins when agents start handling actual requests. Costs are based on OpenAI’s standard API pricing model.

n8n, features and a little complexity

We have also developed an agentic workflow on n8n which allows emails to be read directly from a chat.

To do this, the workflow we want to create must allow the user to request a summary of their latest emails via chat. The workflow consists of six steps. The “When chat message received” node plays the role of the trigger. It waits for a user to send a message in the chat. The setting is: Response Mode > Using Response nodes. This allows you to control what is sent back to the chat.

Then, the “Get many messages” node (from Gmail) connects to the mailbox and retrieves in our example the last 3 emails received. The settings are: Resource > message and Operation > Get many.

After it, the “Edit Files” node cleans and organizes the email data. For this, in the “message” option, we wrote: “You are an assistant responsible for summarizing emails. Here are the emails to process: {{ $json.message }} Summarize these emails in a clear and concise manner.”

The “Code in Javascript” node combines the different emails into a single text. It is set to Mode > Run Once For all items

The code written is:

// Combiner les emails en un seul texte formaté
const emails = items.map((item, index) => {
  const from = item.json.From || 'Non spécifié';
  const subject = item.json.Subject || 'Pas de sujet';
  const content = item.json.message || item.json.snippet || '';
  return `Email ${index + 1}:nExpéditeur: ${from}nSujet: ${subject}nContenu: ${content}`;
}).join('nn---nn');
// Retourner le résultat
return ({
  json: {
            allEmails: emails,
            chatInput: `Voici les 3 derniers emails reçus. Résume-les de manière claire et concise :nn${emails}`
  }
});

Following this, the Basic LLM Chain node, associated with OpenAI Chat Model, is the “brain” of the workflow. He understands the content and must create an intelligent response. The settings are: Connected chat trigger mode” and the prompt: fx {{ $json.chatInput }}

The last node, “Respond to Chat”, returns the in-chat summary to the user.

Once these settings have been made, we can ask the chat to summarize the last 3 emails received.

The agent responds correctly. Overall execution time is around 6 seconds.

For feedback, in terms of ease of use, we see that the visual interface of n8n is a great help. It allows you to better understand the flow. The inputs and outputs of each node make debugging easier. Expressions underlined in red alert you to problems, and once understood, the logic becomes clearer. The AI ​​assistant can also help solve a number of problems. On the other hand, building a workflow on n8n requires a certain number of concepts to master. Expressions in n8n require understanding the data structure. The number of nodes can also make you hesitate about which one to take. Error messages are not always easy to decipher for newbies.

There are no complaints about the wealth of options. Native AI integration features LangChain, agents, and onboard memory. Many connections are possible, such as Gmail and OpenAI for our workflow. The code can be written in JavaScript or Python. But this abundance also brings the weakness of n8n, at least for beginners. Given the number of options available, the documentation is sometimes insufficient for the cases treated.

Remember that n8n offers a free version and starts from €20 per month for 2,500 executions.

Verdict: n8n more complete, while waiting for AgentBuilder to expand