How I AI: 4 Hands-Free Workflows Using Codex Browser Automation
I walk through my daily hands-free AI workflows, showing you exactly how I use Codex browser and computer use to QA my app, conduct persona-based research, manage my LinkedIn inbox, and even go shopping.
Claire Vo
One of my favorite guests ever, Jesse Genet, said on the show that AI was great for moms because our hands are always busy. We’re always doing something, and we don’t have time to be tied to a keyboard. She was talking about voice and chat, but in this episode, I’m taking it a step further. I’m calling this one “Look, ma, no hands,” because we’re diving into browser and computer use—letting AI navigate your computer and websites to get work done for you.
This technology has become exceptionally good recently, especially with what I’ve seen from Codex (what we're now calling the ChatGPT desktop app) and the latest models. Think of it like having a virtual coworker who can literally put their hands on your keyboard and get things done. While that might sound a bit scary, it has been one of the most life-changing AI features I've adopted. It's reduced so much personal and professional digital toil that I had to share the specific workflows I use every single minute of every day.
In this post, I'll walk you through a builder use case for testing software, a productivity hack for managing my inbox, and a fun personal one that got me ready for my vacation. The trick, as always, is finding the right use case. My hope is that these examples will give you some ideas for how you can reclaim your time and let your AI take the keyboard for a while.
Workflow 1: Automated QA Testing for a Web App
As a builder, one of the most practical applications for browser automation is quality assurance. I can write backend tests all day, but what I really need to know is: is my application usable? This is where I lean on Codex to act as a tireless QA tester for my product, ChatPRD.
The Setup and The Prompt
The process is pretty simple. First, you need the desktop app of your chosen model, like the Codex (ChatGPT) desktop app or the Claude desktop app, and its corresponding Chrome extension installed. This combination allows the AI to both see and control your browser.
For this test, I opened my local development version of the ChatPRD onboarding flow. My goal was to validate the user experience, check for mobile responsiveness, and log any issues. Here’s the prompt I gave it:
@browser to test the open onboarding flow for usability and mobile responsiveness. Please screenshot as you go, make a Google spreadsheet of any issues you identify.
Invoking @browser tells the model to use the browser window embedded within the desktop app. You can also use @Chrome to control your main Chrome browser or @computer to give it control over your entire desktop.
The Results: AI Finds Bugs a Human Wouldn't
What happened next was fascinating to watch. The little glowy cursor started navigating the website, clicking dropdowns, and changing viewport sizes to simulate different devices. The AI is far more exhaustive and organized than I am. When I test things, I admit I mostly do the “happy path” QA—I fill out the forms correctly and click next.
The AI, however, tests for failure states. It immediately found a high-severity bug that had been in the product for a while. If a user didn’t select a required option, they could still click “Continue,” but the app would hang without providing an error message. I had missed this for months because I always selected an option. It also switched to mobile testing, checking for accessibility, overflow issues, and touch targets with a rigor I would never have the patience for.
At the end of the process, it produced a beautiful Google Sheet detailing 11 issues, including the one blocking bug, complete with descriptions, steps to reproduce, and even the screenshots it took along the way. This is an incredibly practical way to improve your product with minimal effort.
Workflow 2: Persona Impersonation for Deeper User Research
While QA is useful, it can be a bit dry. A more creative and insightful workflow I love was actually created by my husband, EJ Lawless. Instead of just testing for bugs, we use browser use to have the AI impersonate specific user personas and give us feedback from their point of view.
The Prompt: Inhabiting the User's Mind
For this, I used Monologue to dictate a more complex prompt. I wanted the AI to go through the core of ChatPRD as three distinct personas: a product manager in a hurry, an engineer turning a PRD into a technical spec, and a team leader checking on usage. I gave it very open-ended instructions.
I want you to go through the core of ChatPRD's chat and document creation mode as three different personas.
1. One as a product manager trying to quickly create a PRD for a feature that they just got out of a meeting about.
2. The second, I want you to impersonate an engineer who is taking that already created PRD and trying to make a technical spec out of it...
3. And the third, as a team leader, try to figure out how your team is using ChatPRD.
...I want you to create a doc providing research style critique about the overall experience, including friction points, places of delight, and improvements you think we can make to serve those three personas better.
The Results: Uncovering Real Friction
This is where things get really interesting. The AI went beyond surface-level feedback. As the engineer persona, it tried to reference the document created by the PM persona. In doing so, it exposed what it called a “structural break.” My app doesn't currently allow for easily referencing documents across different threads. I knew this was a limitation, but seeing the AI hit this wall and articulate its frustration from a user's perspective was incredibly powerful. It felt more real than any synthetic user research report.
It also gave me feedback I didn’t ask for, noting that the loading screen was slow and opaque, which was especially helpful because it encountered an error during its run. This persona-based approach gives you fresh eyes—or “agent eyes”—on your product, revealing friction points that you, the builder, have become blind to.
Workflow 3: Hands-Free LinkedIn Inbox Management
Moving on to a productivity use case. I love connecting with all of you on LinkedIn, but my inbox gets overwhelming. Since LinkedIn doesn't have an official API for this, browser use is the perfect solution for getting through the mountain of messages.
The Prompt: A Simple Triage System
I log into LinkedIn in the Codex sidebar and give it a straightforward set of instructions to act as my assistant.
Please use browser to go through my recent unread messages and reply to any that are super critical for ChatPRD or the podcast. Leave me notes here on the others about how I should reply. If this is just a friendly thank you note reply in a friendly way.
The Results: Calibrating for the Right Effort
The AI dutifully clicks through my unread messages, drafts replies for simple notes, and flags the more complex ones for me to handle personally. It’s a huge time-saver.
One key lesson here was about calibrating my model choice. I realized I was using the very powerful GPT-5.6 Soul on high effort for this task, which was total overkill. It was like using a sledgehammer to crack a nut. I was able to switch to a less powerful, medium-effort model (Tara) which was much faster and cheaper, without any drop in quality for this specific task. Matching the model's intelligence level to the job at hand is crucial for making these workflows efficient.
Workflow 4: The AI Personal Shopper
Finally, let's talk about a fun personal use case: shopping. I was heading to Hawaii and needed some new clothes, but didn't have time to browse. Free People was having a sale, so I put my AI personal shopper to work.
The Prompt: A Highly Specific Shopping List
This time, I used the @Chrome command to have the AI work in my main browser window. I gave it a very specific brief, including personal context that would help it make good choices.
@Chrome I'm going to Hawaii this weekend and I need new clothes, good news, exclamation mark. Free people has 25% off their sale. I would love for you to go through using Chrome browser use and pick out 10 items in my size, you can pick size, medium. That would be great for Hawaii. Remember, I'm a breastfeeding mom. I just had a baby six months ago. So pick things that are comfortable, that are breastfeeding friendly, and that would be great for the weather in Hawaii in late July. what I want you to do is end up with 10 things in my cart that I can pick from and buy. Thanks.
The Results: A Curated Cart (With a Little Help)
I love that the Codex app shows a small window of what the browser is doing, so I could monitor its progress. It navigated to the sale section, filtered by size, and started analyzing items. It did hit one snag—the website detected it as a bot and threw up a CAPTCHA. This is a great example of the human-in-the-loop dynamic. I stepped in, solved the puzzle, and handed control back to the AI.
I don’t see this as a failure. It's the proper division of labor: the AI does the hours of tedious browsing and filtering, and I step in for the one minute that requires a verified human. In the end, I had a cart with 10 items—maxi dresses, loose pants, and button-down shirts—all well-suited for my trip. It did a fantastic job.
More Rapid-Fire Use Cases
These four workflows are just the beginning. I use computer and browser use for so many other things:
- Filling out annoying forms: Registering my kid for camp or filling out procurement documents.
- Controlling my phone: I used Codex with Apple’s iPhone mirroring feature to remotely access my home Wi-Fi router's app from a hotel room. The AI opened firewall ports so I could SSH into my Mac Minis, and then closed them when I was done. It was incredible.
- Creating Google Docs and Sheets: When a built-in tool or plugin fails, I just tell the AI, “Please just open my browser and do it in the browser.” It almost always works.
Browser and computer use is genuinely my favorite AI feature right now. It's changing how I interface with my digital life, automating the toil and freeing me up to focus on what matters—or to just not be on my computer at all. The goal for me is to be able to enjoy that Hawaii vacation, knowing that things are still getting done back home.
I'd love to hear how you're using these features. What workflows have you built? What questions do you have about privacy or security? Let me know in the comments.
---
Thanks to Our Sponsors
Thank you to the sponsors of 'How I AI' for their support:
- Runway: The creative AI platform for images, video and more
- Hyperagent: Deploy fleets of agents that handle real work
