AI Folks
  • Courses
  • Web Of AI
  • Pricing
  • Enterprise
Log in
PreviousBuild with Google AI StudioNextThe OpenAI Agents SDK

Vibe Coder · Unit 16: AI Agents · Lesson 5 of 9

Choosing an agent builder

Agent tools appear and disappear. OpenAI is shutting down its visual Agent Builder. How to choose a builder, and how to keep your agent portable when one goes away.

Read the short notes and pictures.Do everything inside a blue frame.

A builder you cannot count on

In 2025 OpenAI released Agent Builder, a visual canvas for multi-step agents: you dragged nodes such as Agent, File search, Guardrails, MCP, If/else and Human approval onto a canvas, previewed runs and published a version. OpenAI’s documentation now states that Agent Builder is being deprecated and is scheduled to shut down on 30 November 2026. ChatKit, the chat interface that went with it, remains available, and the code route is the Agents SDK.

A chalkboard, hand-drawn. A table: the visual canvas is being deprecated, shutdown on 30 November 2026, ChatKit remains, the Agents SDK is the code route.
OpenAI’s own docs give the shutdown date.

This course does not teach Agent Builder

There is no point learning a tool that closes this month. The ideas in it, an agent node, file search, guardrails, human approval, are the same ideas you are learning in n8n and Langflow, and they carry over. Engineers will meet the Agents SDK later in this unit.

What you own

When a builder goes away, the canvas goes with it. What you keep is whatever you stored outside it. Three things define an agent, and all three can live in plain files.

A chalkboard, hand-drawn. Three cards: instructions, tool definitions and tests.
These three define the agent, whichever tool runs it.

Instructions

The system prompt, saved as a text file and changed through commits.

Tool definitions

For each tool: its name, description, inputs and what it returns.

Tests

A list of real questions with what a good answer contains. This is how you know a rebuild works.

Locked in

The prompt lives only in a node.
Tools are clicks on a canvas.
No test set.
Moving means starting again.

Portable

agent/instructions.md
agent/tools.json
agent/tests.md
Moving means rebuilding from files.
The second column takes ten minutes to set up and is worth doing on day one.

Choosing

  1. 1

    Can I export?

    A workflow file, a flow file, or code. If nothing leaves the tool, be careful.

  2. 2

    Can I change the model?

    Prices and quality shift every few months. You want to be able to switch.

  3. 3

    Where does it run and who pays?

    Hosted for you, or on your own server. Per task, per seat or per token.

  4. 4

    Is it maintained and used?

    Recent releases, real documentation, a community. Check the docs for a deprecation notice.

n8n

When the agent is one step in a workflow with triggers and integrations.

Langflow

When the app is mostly the model, with retrieval and tools, and you can host it.

Google AI Studio

When you want a small Gemini app live quickly.

An agent SDK

When an engineer needs full control, tests and version history in code.

ActivityYour turnWrite it yourself first, then compare.

A new agent builder launches next month and looks excellent. Write the three checks you would make before moving your product’s agent to it.

ActivityBuild stepAn agent that works for your product

Make your agent portable

In your product’s repository, create an agent/ folder with three files: the instructions, the tool definitions and your test questions with what a good answer contains. Add the exported n8n workflow and Langflow flow beside them, then commit and push.

Done when

From now on, change the instructions in the file first and then in the tool, not the other way round.

Keep these

  • OpenAI’s Agent Builder shuts down on 30 November 2026.
  • Instructions, tool definitions and tests are what you own.
  • Choose builders you can export from and switch models in.
Previous lessonBuild with Google AI StudioNext lessonThe OpenAI Agents SDK