How to Make an Instagram AI Agent Take Actions in ZazzyAgent

A useful AI Agent shouldn't only answer questions.

It should also be able to do something when the conversation requires it.

For example, a customer says:

I'm interested in a demo.

The AI could:

1. Collect the customer's information.

2. Save it.

3. Add a lead label.

4. Start a follow-up.

5. Assign the conversation to sales.

That's what Actions are for.

What is an AI Agent action?

An action tells ZazzyAgent to perform a specific operation after the AI determines that the action should happen.

The current AI Agent system supports actions such as saving custom fields, adding/removing labels, assigning sequences, triggering flows, calling APIs, assigning humans and transferring conversations between agents.

Why actions matter

Without actions:

Customer talks to AI.

With actions:

Customer talks to AI → ZazzyAgent updates the customer record and moves the conversation forward.

That turns an AI chatbot into an operational assistant.

Save customer information

Suppose the customer gives:

My budget is ₹75,000.

Your AI Agent can save that information to the appropriate custom field.

The action follows the syntax:

##save_custom_field## : ID

Use the action autocomplete in the System Prompt instead of guessing the ID.

Add a label

Suppose a customer is a qualified lead.

The agent can add a label using:

##add_label## : ID

For example:

Qualified Lead

Now your team can filter and manage these customers later.

Remove a label

You can also remove an existing label when the customer's status changes:

##remove_label## : ID

For example:

Lead → Customer

Assign a sequence

An AI Agent can start a sequence after the customer reaches a certain stage.

Example:

Customer qualifies.

↓

Add:

Sales Follow-up

The action is:

##assign_sequence## : ID

Trigger another flow

Suppose the AI needs a structured form.

It can trigger a bot flow:

##trigger_bot_flow## : ID

For example:

Customer wants to book a demo.

↓

Trigger:

Demo Booking Flow

This is a powerful way to combine AI conversation with structured automation.

Transfer to another AI Agent

Suppose you use separate Instagram agents:

Lead Agent

Sales Agent

A Lead Agent can transfer the customer after qualification.

The action is:

##transfer_agent## : ID

For example:

Once the lead is qualified, transfer to the Sales Agent.

This is how a multi-agent Instagram setup can work.

Assign a human

When the conversation needs a person, the AI can assign it to your team.

For example:

##assign_human_role## : ID

or:

##assign_human_member## : ID

Use the role action when you want routing among available team members.

Use the member action when the conversation should go to a specific person.

Call an external API

ZazzyAgent can also support API-based actions.

For example, the AI could send information to another system.

The basic action structure is:

##call_http_api## : ID | ...

The exact response behaviour depends on how the API action has been configured.

See HTTP API in ZazzyAgent.

Combine actions

One customer conversation can cause several actions.

Example:

Customer
   ↓
Interested in Demo
   ↓
Collect Name
   ↓
Collect Email
   ↓
Save Custom Fields
   ↓
Add Lead Label
   ↓
Trigger Booking Flow

The important part is to make sure the actions happen in the correct order.

Use actions conditionally

Don't tell an AI:

Always add Qualified Lead.

Instead:

Add Qualified Lead only after the required qualification information has been collected and the lead meets the qualification criteria.

This prevents incorrect labels.

Don't use every action

Only give an agent the actions it actually needs.

A lead agent may need:

• Save Custom Field

• Add Label

• Transfer Agent

It may not need:

• Block Subscriber

• Cancel Orders

• Remove Sequences

Narrow action sets generally make agent behaviour easier to control.

Use action autocomplete

When writing a System Prompt, type:

##

in the action area.

ZazzyAgent can show the available actions so you can select the correct one.

This is much safer than manually guessing action names or IDs.

Example Instagram lead agent action logic

If the customer provides their name,
save the name.

If the customer provides their email,
save the email.

If the customer provides their requirement and budget,
add the Qualified Lead label.

Once qualification is complete,
transfer the conversation to the Sales Agent.

This is much better than simply listing actions with no conditions.

Test the actions

Don't only test the AI's written response.

After a test conversation, check:

• Custom fields

• Labels

• Sequences

• Flow triggers

• Human assignment

• Agent transfer

• API results

An AI saying:

I've saved your details.

doesn't prove that the action actually happened.

Verify the underlying customer record.

The action architecture

A useful mental model is:

Instagram DM
     ↓
AI understands request
     ↓
Decision
     ↓
Action
     ↓
Customer / System updated
     ↓
AI continues conversation

Actions are what connect the conversation to the rest of your business process.

zazzyagent.com