{"id":31818,"date":"2025-10-13T12:00:00","date_gmt":"2025-10-13T10:00:00","guid":{"rendered":"https:\/\/salvadorvilalta.com\/?p=31818"},"modified":"2025-10-10T18:34:26","modified_gmt":"2025-10-10T16:34:26","slug":"agentkit-the-canva-of-ai-agents-is-now-real","status":"publish","type":"post","link":"https:\/\/salvadorvilalta.com\/en\/agentkit-the-canva-of-ai-agents-is-now-real\/","title":{"rendered":"AgentKit: the ‘Canva’ of AI agents is now real”."},"content":{"rendered":"\t\t
\n\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

For years, talking about creating artificial intelligence agents was synonymous with complex things: lines and lines of code, APIs, endless prompts, models that nobody fully understood… and more than one frustration along the way.<\/p>

But what if you could create your own intelligent assistant by simply dragging and dropping blocks on a screen, as if you were making a presentation or a mind map?<\/p>

That’s just what OpenAI is proposing with Agent Builder<\/strong>, its new visual tool for designing AI agents without the need to know how to program. And yes, when Sam Altman compared it to Canva, he wasn’t exaggerating. <\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t

\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

What exactly is Agent Builder?<\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Agent Builder is part of the new \n AgentKit<\/strong>\n<\/a>a sort of “toolbox” for creating, testing, and deploying custom intelligent agents. But the most striking thing of all is this: you don’t need to write code<\/strong> to make it work. <\/p>

With Agent Builder you have a visual environment where you can build the logic of an agent by simply connecting nodes that represent steps: a decision, a search, a classification, a response… all in a graphical, clear and editable way.<\/p>

And the best part: you can use predefined templates (like a travel assistant or a customer service bot) or start from scratch if you’re more into customization.<\/p>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t

\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t
\n\t\t\t
<\/div>\t\t<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

How does it work in practice?<\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Imagine you want to create an agent to help your customers manage orders.<\/p>

  1. You start with a start node<\/strong>, where the system receives the query.<\/p><\/li>

  2. Then you place a sorting agent<\/strong> that analyzes what the user wants (shipment tracking? returns? product inquiry?).<\/p><\/li>

  3. You then use a conditional node<\/strong> to branch the flow according to the type of query.<\/p><\/li>

  4. You add specific nodes for each type of task: for example, one that connects to your database to get the status of the order, and another that returns steps to make a return.<\/p><\/li>

  5. If you are worried about the system saying something out of line, you can add a guardrail <\/strong>(a kind of safety check) to filter or block certain responses.<\/p><\/li><\/ol>

    And all this without writing a single line of code. Literal. <\/div>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
    \n\t\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t\t\t\t\t\t\t\t\t
    \n\t\t\t\t\t\t\t\t\t\t\"OpenAi\t\t\t\t\t\t\t\t\t\t\t
    Source: OpenAI<\/figcaption>\n\t\t\t\t\t\t\t\t\t\t<\/figure>\n\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
    \n\t\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t\t

    And what happens next?<\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
    \n\t\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t\t\t\t\t\t

    Once you have the flow assembled and tested, you can publish it and use it in various ways:<\/p>