How to use Action Agent (BETA)

Who can use this feature

Supported on Starter, Team and Enterprise plans

We’re launching Action Agent, a groundbreaking beta feature, and it’s not just another tool—it’s an autonomous AI teammate ready to do the work for you!

Imagine describing a final goal, and that’s it. No more step-by-step instructions. This feature takes your request and independently builds the solution, whether it's a detailed report, a complex data analysis, or a fully-functional webpage built from scratch.

This is a monumental leap in productivity, turning your biggest ideas into finished products in a fraction of the time. We’re giving you the power to accomplish more, faster than ever before.

This is a really big deal, and we want you to be among the first to experience it.

Latest updates

What is Action Agent?

The best way to understand Action Agent is to think of it as a digital teammate. Like any great teammate, it's collaborative, capable, and ready to take on tasks. It has a wide range of skills, from research and writing to data analysis and coding. Its goal is to take work off your plate so you can focus on the bigger picture.

Beyond answering questions

You've probably used tools that can answer questions, like "What is the capital of Australia?" or "Summarize the plot of Hamlet." These tools are great for finding information. Action Agent is different because it uses information to complete tasks.

Instead of asking it to tell you how to make a presentation, you can ask it to make the presentation for you. It will find the information, structure the slides, write the content, and deliver a finished file.

From goal to finished product

The most powerful aspect of Action Agent is its ability to work autonomously toward a goal. You don't need to know every step required to get something done.

You say: "Analyze the attached sales data and create a report on our top five performing products this quarter."

Action Agent does this: Opens the file, analyzes the data, identifies the top products, writes a summary of the findings, formats it into a clean report, and gives the final document to you.

You provide the destination, and Action Agent figures out the best way to get there.

How to access Action Agent

Select Action Agent from the left sidebar menu, or select the direct link here.

You can also find Action Agent in our Agent Library.

Navigating Action Agent

Similarly to Ask Writer, you’ll have sessions displayed on the left side. Selecting the + icon creates a new session.

Select the paperclip icon to add source files. Action Agent can handle just about any file type - if it isn’t able to handle a type of file, it has the ability to convert it or process it in a way it can use! ⚡

Select the microphone icon to enable voice-to-text and speak directly to Writer.

Enter your prompt, then select enter to begin.

Below the input field, we offer some suggested use cases. Select one to load the prompt into Action Agent and see the full range of its capabilities!

Once you’ve submitted a prompt, a new window will appear on the right side. In the center pane, you’ll see Action Agent keep a running explanation of everything its doing, and why its doing it. On the right side, you can see a status update for each step being conducted by Writer’s computer. You can use the scroll bar at the bottom of the panel to view previous steps in the process.

Note: You can interrupt Action Agent at any time. If you see a fault in its logic, or simply want to change course from your original action plan, just send a followup message in the central panel.

When Action Agent has produced the final output(s), you’ll see them displayed on the right side. You can select the files to expand them, download them, open them in a new window, etc.

Note: You can navigate away from Action Agent without interrupting your workflow. It will continue to run as long as it needs to complete its assignment.

How does Action Agent work?

Working with Action Agent is a simple, transparent process. You’re always in control and can see exactly what it’s doing.

Step 1: Make a request.

Begin with a clear, goal-oriented request. The more specific you are about what you want the final product to be, the better. For example, instead of “Tell me about our sales,” a better request would be, “Create a 10-slide presentation summarizing our Q3 sales performance, using the data in sales_q3.csv.”

Step 2: Action Agent creates a plan

Before it does anything, Action Agent will outline its plan to achieve your goal. It will show you a checklist of the steps it intends to take. This gives you a chance to review its approach and ensure it’s on the right track before the work begins.

Step 3: Action Agent gets to work

Once it has a plan, Action Agent starts executing it one step at a time. It has a full toolkit at its disposal, allowing it to perform tasks like:

Browsing the web for information
Reading and writing files (like documents, spreadsheets, and code)
Analyzing data to find insights
Creating new documents, presentations, and spreadsheets
Writing and running code to build applications or perform complex calculations

Step 4: It delivers the result

When all the steps in its plan are complete, Action Agent will present you with the finished product. This might be a Word document, a PowerPoint presentation, an Excel spreadsheet, or even a zip file containing a fully functional web page. You can then review the work and ask for revisions if needed.

Tips for getting the best results

To make your experience with Action Agent as effective as possible, keep these tips in mind:

Be clear about your goal: The more specific you are about the final product, the better. Instead of "make a presentation," try "make a 10-slide presentation for a sales kickoff meeting."
Provide context: If you have relevant files, data, or links, attach them to your request. The more information it has, the more accurate its work will be.
Think big: Don't be afraid to give it complex, multi-step tasks. Action Agent excels at breaking down big projects into manageable steps. A request like "Research our top three competitors and create a competitive analysis presentation" is a perfect task for it.
Review the plan: Pay attention to the plan it creates. If it doesn't look right, you can ask it to adjust its approach before it starts the work.

Example prompts & use cases

Just ask Action Agent:

Use Case: Understand the scope of potential projects handled by Action Agent.
- Example Prompt: "I am a {sales engineer} at {Brickly.ai}. Generate a list of robust use cases for Action Agent, with example prompts. These should be complex, second-level tasks that deliver high quality results and take advantage of Action Agents tool calling, orchestration and reasoning skills."

Financial Services

Use Case: Prepare a comprehensive pre-meeting briefing package for a high-value potential client.
- Example Prompt: "I'm meeting with the CFO of 'Global Tech Innovators Inc.' next week to discuss our wealth management services. I need to be fully prepared. Can you search the web for their latest financial reports, news, and strategic initiatives? Also, pull together a profile on the CFO. Based on that, can you identify a few potential financial challenges or opportunities they might have? Please synthesize everything into a structured briefing document with sections for a company overview, key financials, the CFO's profile, and some potential talking points for me."
Use Case: Create a multi-channel marketing campaign for a new investment product.
- Example Prompt: "We're launching a new 'Future-First ESG Fund' for tech-savvy millennials. I need a campaign to get the word out. Could you start by researching the financial concerns and communication styles of this audience? Then, write a 1,200-word blog post titled 'Why Your Portfolio Needs to Be as Forward-Thinking as You Are,' a script for a 60-second promo video, and a few engaging posts for LinkedIn. Also, could you create the HTML/CSS for a simple landing page to capture leads? Finally, package all of it into a zip file for me."
Use Case: Design and document a more efficient client onboarding workflow.
- Example Prompt: "Our current client onboarding process is too manual. I want to streamline it. Can you help me design a new workflow? Please create a flowchart using Mermaid syntax that shows the new, more automated process. Then, write up a formal Standard Operating Procedure (SOP) based on that flowchart, and also create a simple, friendly PDF checklist we can give to new clients to guide them through the steps."

Retail & Consumer Packaged Goods (CPG)

Use Case: Develop a new product.
- Example Prompt: "I am a researcher at ChocoFlava Corp. I am trying to combat declining sales by coming out with new flavors and compositions of candy. I’d like to build a comprehensive end-to-end plan where we research what trends and competitors are rising, and then create an idea for a new candy bar or candy product, and then come up with a name, marketing plan, customer personas, etc. Let’s first focus on the competitive landscape, then map out the ideas, personas, and plan."
Use Case: Create customized sales pitches for different retail channels.
- Example Prompt: "I need to pitch our new 'Verve' protein bar to two very different retailers: a high-end grocery chain and a chain of fitness centers. Could you help me create a tailored sales deck for each? For the grocery chain, the pitch should focus on organic ingredients and clean-label trends. For the gyms, it should emphasize protein content and post-workout benefits. I'll need two separate 5-slide PowerPoint presentations, each with a slightly different angle."

Technology

Use Case: Develop a research plan for a product in development.
- Example Prompt: "I’m working on a new product that aims to add memory to chatbots powered by large language models. Can you search Hugging Face and find me the 10 most popular research papers on the topic of providing LLMs with memory published in the last six months. Create a CSV table with the title, link to the paper, and name of the lead or corresponding author. Then search the web and see if you can find email contacts or personal websites for each of these authors. Finally, give me a quick written breakdown of the tools and techniques these researchers used to add memory to LLMs and rank them in order of complexity."
Use Case: Transform a creative brief into a project plan and generate supporting assets.
- Example Prompt: "You are a marketing program manager coordinating a new campaign. Start with the creative brief and brand voice guidelines and generate a marketing copy and design in Google Docs. Create a working plan that outlines work which is needed to launch this campaign, and assign teams that are appropriate to carry out this work. If tasks are missing owners or dependencies aren’t clear, flag them for review rather than guessing. Output should be a fully launched campaign with tracked links and live assets."
Use Case: Enrich inbound leads, assign prioritization, and route to reps.
- Example Prompt: "You are a RevOps assistant qualifying new inbound leads.
- Enrich the lead using ZoomInfo, Check if the lead meets ICP criteria using internal documentation. Assign a prioritization level and route the lead based on the current routing rules. Output should be a fully routed lead record in Salesforce."

Healthcare & Life Sciences

Use Case: Prepare a comprehensive package for a hospital's Value Analysis Committee (VAC) meeting.
- Example Prompt: "We have a critical VAC meeting at St. Jude's Hospital to get our new 'CardioView 2.0' device approved. I need the ultimate prep package. Can you create a cost-benefit analysis in Excel, modeling the potential savings for the hospital based on the attached data? Also, please review the hospital's procurement guidelines (attached) and create a checklist showing how our device meets their requirements. Finally, could you write three short, powerful testimonial quotes from fictional doctors who love the device?"
Use Case: Streamline the clinical trial documentation submission process.
- Example Prompt: "Our process for submitting trial documentation to regulatory bodies is a mess. I need to standardize it. Can you start by creating a master checklist in a spreadsheet of all the documents required for an FDA submission? Then, create Word templates for three of our key documents: a Protocol Summary, a Patient Consent Form, and a Site Information Sheet. Finally, please design a Mermaid flowchart that visualizes the entire process, from creation to final submission."
Use Case: Identify ideal clinical trial locations.
- Example Prompt: "Please identify optimal locations for a new clinical trial of a drug to treat eczema? Please generate a ranked list of recommended trial sites, complete with a detailed report including maps and charts for the clinical operations team. Use factors like demographics, affordability, available space, and local regulations to rank the candidate site."

How can I provide feedback on Action Agent?

We want your feedback! When Action Agent produces an output, please let us know how it did by selecting Nailed it or Missed the mark.

You’ll be able to provide additional feedback if desired.

Submitting feedback automatically shares your session with our team.

Frequently asked questions

Tools and capabilities

What tools are available to Action Agent?

Web & Data Interaction:

Web Search: Performs web searches to gather up-to-date information, answer questions directly, and find relevant source URLs.
Web Scraping & Browsing: Scrapes webpage content for detailed analysis and use a full-featured browser to interact with dynamic sites, fill forms, and click elements.
Data Processing: Equipped with tools like jq (for JSON), csvkit (for CSV), and xmlstarlet (for XML) to parse and manipulate structured data.

Software Development & System Operations:

File System: Full capability to create, read, write, modify, and manage files and directories.
Code Execution: Writes and executes code, primarily in Python, and save it to files for reuse and modularity.
Version Control: Uses git to clone, manage, and interact with code repositories.
System Utilities: Can access a standard Linux utility suite (grep, awk, sed, find, wget, curl, zip, etc.) for powerful text processing, file management, and network operations.
Package Management: Installs necessary software packages and dependencies using system package managers.

Deployment & Sharing:

Local Hosting: Runs web servers locally to test applications.
Port Exposure: Use the expose-port tool to make services running in its environment publicly accessible via a URL, allowing users to interact with its work (e.g., a web app I've built).
Static Deployment: Deploy static websites (HTML/CSS/JS) to a production environment.

Security, transparency & other guardrails

How do we mitigate security, safety, and privacy risks given the level of autonomy and access of Action Agent?

Action Agent is designed with a multi-layered approach to mitigate risks associated with its autonomy, and backed by the security, reliability, and control provided by the WRITER platform.

Security Controls:

Isolated Sandbox Environment: Operates within a secure, containerized sandbox that is completely isolated from the host infrastructure and other user sessions.
Temporary Workspace: All files, data, and running processes are temporary and are permanently destroyed at the end of each session. This design prevents the persistence of malware or sensitive data.
Restricted Network Access: Outbound network connections are monitored and filtered, limiting it to standard web protocols (HTTP/S) and preventing malicious activities like network scanning or participation in DDoS attacks.

Safety Protocols:

Behavioral Guardrails: The underlying language model has built-in safety filters to prevent the generation of harmful, illegal, or unethical content.
Human-in-the-Loop Design: For critical actions, such as deploying applications or deleting files, the agent is programmed to stop and request explicit confirmation from the user. The user retains ultimate control over significant operations.
Resource Limiting: All processes are strictly limited in their use of CPU, memory, and execution time to prevent system instability or resource abuse.

Privacy Protections:

Stateless Sessions: Action Agent has no long-term memory and does not retain any data between sessions. Each interaction begins from a clean slate.
User-Controlled Data: Action Agent only has access to data that the user explicitly provides during an active session. It has no access to a user's local machine or files.
Encrypted Communication: All data transmitted between the user and the agent is protected with industry-standard HTTPS encryption.