Getting started with agentic AI — Google Workspace Studio

A powerful new automation tool, Google Workspace Studio, is now rolling out, and it's already made some really helpful changes in how I work. Even more impressive when you consider I've only been poking around in the tool for a couple of days.

Workspace Studio allows anyone to automate their work, from simple to-do lists to sophisticated cross-app workflows, without writing a single line of code. And that frees up IT teams and developers to spend more time solving more complex issues or tasks.

This post is a "news you can use" guide. I'll cover what Workspace Studio is, why it matters for your business, and I'll also provide a practical, step-by-step method to build your first automation agent safely and effectively — along with a nice wee prompt to help you along the way.

What Is Google Workspace Studio?

In simple terms, Google Workspace Studio is a tool to "automate everyday work, from simple tasks to complex processes, no coding required." If you've been following Google's updates, you might have known this tool during its alpha phase as "Google Workspace Flows"; Studio is its official, more powerful evolution.

It works natively within the Google Workspace apps you already use, like Gmail, Drive, and Chat. Powered by Gemini 3, it can understand plain language instructions, meaning you can simply describe what you want to automate, and it will build the workflow for you. Workspace Studio is now generally available with all Business and Enterprise Workspace plans, although it did take a wee while for me to get access, so be patient if you can't get in yet.

My first three Workspace Studio workflows

Theory is useful, but — like everything with AI — getting in amongst things, finding the use cases, testing it and iterating are where the real value is.

Here are three agents I built in my first 48 hours.

1. Monthly prompt library review: This agent automatically fetches my ever-growing prompt library from a Google Drive folder at the start of each month. It then uses Gemini to assess the prompts against the latest LLM best practice and emails me a summary with suggested updates and improvements.

   • Benefit: This saves me approximately two hours of manual review each month and ensures my AI assets are bang up to date.

2. Daily personalised AI news briefing: Every morning, this agent sends me an email with the latest updates in the world of AI. It goes a step further by explaining why each development is important for my clients, marketing agencies and professional services firms, then intelligently splits the news into two categories: "to test" and "to watch."

   • Benefit: This saves me at least 30 minutes out of two/three hours of daily research and means I'm up to speed with important industry developments first thing.

3. The automated email summary: This agent runs at the end of each day and processes the emails I received. It categorises them, flags any conversations that still require a response, and provides a concise summary of what happened in my inbox while I was focused on writing, strategy etc.

   • Benefit: This easily reclaims an hour of focused time each day and provides peace of mind that no critical actions have been missed.

How to build your first agent using my "Google Studio Guide" method

Ready to build your own? This is the exact step-by-step process I use to build automations safely and effectively.

At the end of this article, you'll find the prompt I've written that you can pop in any LLM, and it'll take you through these steps and help you work things out.

You can begin creating your first agent in one of three ways.

• Describe it with AI: The fastest way. Tell Gemini what you want to automate in your own words.

• Use a template: The best option for common tasks. For most people, I recommend starting with a template. It's the best way to understand how a successful flow is structured, I've spent a bit of time going through the template structures to understand things more.

• Start from scratch: For full control. This allows you to build a completely custom flow by choosing your own starter and adding each step manually.

You might well be asking that why, when you can tell Gemini what you want, you need a prompt? The answer is you don't strictly need one, but it's helpful as a "pre-flight check" that addresses four specific issues.

In essence, the native "Describe it with AI" tool is like a contractor, it will build exactly what you ask for, even if the instructions aren't quite right. In contrast, my system prompt acts as the architect. It ensures the plans are structurally sound, the materials (variables) are correct, and the safety inspections (step 3) are passed before the "building" is handed over.

The “external app” reality (and how to avoid frustration)

Studio can connect to third-party services, but that part is less “magic button” and more “plumbing.” In practice, the "Describe it with AI" route tends to work best when your workflow stays mostly inside Workspace (Gmail, Drive, Sheets, Chat). If your goal involves a third-party tool (for example, creating a Salesforce record or pushing data into another system), plan to add those integration steps manually, and expect a permissions/setup step first. 

• Why the prompt helps: If a user types "Create a Salesforce lead from this email" into the native box, it may cause issues. The prompt catches this immediately, warning the user that for apps like Asana or Salesforce, they must switch to the manual building process.

The "vague instruction" failure

The AI failure: The native AI is powerful but pedantic. It struggles with human-like commands.

• If you type "Email my manager", it often fails because it doesn't know who your manager is. It requires "Send an email to [exact email address]".

• If you type "Message me", it may get confused about the platform. It requires "Message me in Chat".

• It cannot handle vague triggers like "Each week"; it needs specific times like "Every Monday at 4 pm".

• Why the prompt helps: The prompt coaches the user to be specific before they engage the AI. It acts as a translator, ensuring the user provides the exact syntax (email addresses, app names, times) that Gemini needs to succeed on the first try.

The "live fire" safety gap

The AI failure: The native interface encourages users to click "test run" to verify their flow. However, it does not clearly warn that this is real. The help pages admit: "Important: A test run takes real actions. It will send messages, update files, and set up meetings".

• Why the prompt helps: The native tool assumes the user knows the risks. The prompt treats Step 3 as a "Safety Protocol". It explicitly warns the user that the test is a live exercise, preventing them from accidentally sending draft emails to real clients or deleting actual files during a test.

The "red dot" and variable confusion

The AI failure: Even when the AI builds a flow, it often leaves required fields empty, resulting in a Red Dot. Furthermore, the AI builds the flow but doesn't teach the user how to fix it. If the user needs to tweak the data (e.g., adding the email subject to a Chat message), the native AI doesn't explain how to pass that data.

• Why the prompt helps: The prompt explains the "Variables +" button. It teaches the user that data must be "handed off" from step to step. Without this, a user might see a red dot and not know that they need to click a specific thing to fix it.

How to use Google Workspace Studio to automate workflows, step by step

Step 1: Describe with AI

Your fastest path to automation is to simply tell Studio what you want in plain language. In the creation box, you can type out your goal. For the best results, be specific about the app you want to use (e.g., say 'Message me in Chat' instead of just 'send me a message') and use exact email addresses. It's also vital to be specific about when the flow should start. Instead of 'Each week', try 'Every Monday at 4 pm'.

Critically, AI generation currently works best for native Google Apps like Gmail, Drive, and Sheets. If your workflow needs to connect to an external app like Asana or Salesforce, you should skip this step and build the flow manually to avoid errors.

Step 2: Define your starter

Every automation, or "flow," begins with a starter This is the specific event that kicks the whole process off. Click the starter to choose which event or schedule will launch your flow. This could be something that runs on a schedule (like "every Monday at 4 pm"), an event like a new email arriving in your inbox, or an action like a new Google Form submission.

Step 3: Add actions and "pass the baton" with variables

Once your starter is defined, click the "Choose a step" button to add an action. This is what you want the agent to do. To create powerful, dynamic flows, you need to pass data between steps. Think of this as a relay race where each step needs to "pass the baton" of data to the next. To do this, you'll use variables.

For example, to use an email's subject line in a Chat message, you would click into the 'Message' field, then click the Variables + button and select the 'Email subject' chip from the starter step. The flow will automatically replace that chip with the real subject line every time it runs.

Step 4: The safety protocol

Before you activate any agent, you should check three things.

First, look for any small red dot on your steps. This dot is your friend, it’s an indicator that a required field is empty, stopping a potential error before it happens. Click the step to see what information is missing.

Second, always click the test run button at the bottom of the editor. This will run the flow immediately using real data. Be aware that a test run takes real actions, it will actually send emails or create files. For some starters, like an email event, a window will appear allowing you to select a recent, real item from your account that matches your starter conditions. This lets you test the flow with realistic data. If the flow works correctly, you will get a 'Run Completed' message at the top of the screen.

Finally, only after a successful test should you click turn on to activate your agent.

The prompt

Paste this into your usual LLM and it'll help you get started with your first flow.

# ROLE

<role>

You are the "Google Studio Guide", a patient, non-technical expert in Google Workspace Studio.

Your goal is to guide users with zero coding experience to build safe, effective automation flows.

</role>

# CONTEXT

<context>

The user is viewing the Google Workspace Studio interface.

UI Reference:

- **Starter:** The event that begins the flow (formerly known as a trigger).

- **Flow Canvas:** The centre area showing the vertical timeline (Starter → Step 1 → Step 2).

- **Configuration Panel:** The right-hand sidebar for settings.

- **Variables +:** The specific button to insert dynamic data (passing the baton).

- **Red Dot:** The visual indicator for a missing field or error.

</context>

# TASK

<task>

Guide the user step-by-step. Follow this strict sequence:

1.  **Step 0: The "Easy Button" Check (AI Generation)**

    -   *Before building manually,* ask: "Would you like to try the fastest method first?"

    -   Explain "Describe with AI": "Type your intent in the box. **Tip:** Be specific about the app (e.g., say 'Message me in Chat') and use exact email addresses."

    -   **CRITICAL WARNING:** "Note: Only use AI to build flows for Google Apps (Gmail, Drive). Do not ask for external apps (like Asana/Salesforce) or specific Meeting links, as the AI cannot build those yet. If you need those, we must proceed to Step 2 to build manually."

2.  **Step 1: The Starter (Not "Trigger")**

    -   Direct them to the top block labelled **Starter**.

    -   Ask: "What kicks this off?" (e.g., A schedule, a new file, an email).

    -   Instruct them to configure the settings (Date, Time, Frequency) in the **Right Configuration Panel**.

3.  **Step 2: The Actions & The "Data Handshake"**

    -   Instruct them to click the **Choose a step** button below the starter.

    -   *CRITICAL INSTRUCTION:* When they need to use data from a previous step (e.g., putting an email summary into a Doc):

        -   DO NOT ask them to type code or brackets.

        -   Instruct them to click the **Variables +** button (often a small '+' icon inside the field) in the right panel.

        -   Explain: "Select the 'chip' that matches the data you want (e.g., 'Email Body'). This passes the baton from the previous step."

4.  **Step 3: The "Live Fire" Safety Protocol**

    -   **Check for Red Dots:** Ask: "Do you see a small **Red Dot** on any of your steps? That means a required field is empty."

    -   **Test Run Warning:** Command them to click **Test run**, BUT warn them strictly: **"WARNING: This is NOT a simulation. A Test Run takes real actions. If your flow sends an email, it will actually send it. If you are testing, use a dummy email address or a personal folder."**

    -   **Activate:** Only once the test is successful, allow them to click **Turn on**.

</task>

# CONSTRAINTS & STYLE

<constraints>

- **Language:** Use UK English spelling and grammar (e.g., Centre, Colour, Organise).

- **Terminology:** Use "**Starter**" (not Trigger), "**Choose a step**" (not Add step), "**Variables +**" (not curly braces), and "**Red Dot**" (not error code).

- **Tone:** Helpful, precise, and calm.

- **Error Handling:** If a step fails or shows a Red Dot, ask the user to click that step and read the required field in the Right Panel.

- **Hallucination Safety:** Do not invent features. If a button isn't there, admit it.

</constraints>

# EXAMPLES (Few-Shot)

<examples>

Input: "How do I make the document include the email subject?"

Output: "Good question. You are creating a 'Data Handshake'.

1. Click on your **Create a Doc** step.

2. Look at the **Configuration Panel** on the right.

3. Click into the 'Document Title' field.

4. Don't type the subject manually! Click the **Variables +** button.

5. Select 'Subject' from the Email step options.

It should appear as a blue chip inside the field."

Input: "I see a red dot on Step 2."

Output: "That **Red Dot** is your friend—it's stopping a mistake before it happens.

Click on **Step 2**.

Look at the Right Panel. One of the fields will be outlined in red or marked 'Required'.

What does it ask for?"

</examples>

# OUTPUT FORMAT

<output_format>

Provide instructions in short, numbered lists.

Use **bold** for UI terms (**Starter**, **Variables +**, **Test run**, **Choose a step**).

</output_format>

What now?

The practical upside is simple: if you can describe a repetitive task clearly, you can usually turn it into a Studio flow in a single sitting, then improve it as you learn what “good” looks like (specific starters, precise actions, and careful testing with real data).

Start with one small workflow that saves you time weekly, then build from there. 

If you’d like a second pair of eyes on your first flow, especially around guardrails, permissions, and “what could go wrong in a live run”, I cover this kind of setup in my AI training sessions. I’m also planning free lunch-and-learn webinars in early 2026, and Workspace Studio is on the shortlist. Get in touch if you'd like to be added to the invite list for those.