---
title: 'Parallel AI Agent Browser Automation With Claude Code'
source: 'https://youtube.com/watch?v=cueVwWVoFlU'
video_id: 'cueVwWVoFlU'
date: 2026-07-28
duration_sec: 1045
---

# Parallel AI Agent Browser Automation With Claude Code

> Source: [Parallel AI Agent Browser Automation With Claude Code](https://youtube.com/watch?v=cueVwWVoFlU)

## Summary

This video demonstrates using Claude Code for advanced browser automation, including parallel sub-agents for Amazon product searches, CAPTCHA solving, and automated Reddit account creation with temp mail. The presenter explores how AI agents can perform multiple browser tasks simultaneously and build reusable skills.

### Key Points

- **Parallel Amazon Pipeline** [00:14] — Claude Code spawns four sub-agents to search for Scandinavian-style chair, table, couch, and curtains in parallel, each navigating Amazon independently to find items within a $3,000 budget.
- **Budget and Visualization** [02:51] — The user prompts the agent to find furniture matching a living room image and budget, and the agent uses Nana Banana (with File AI) to generate a visualization of the new room layout.
- **Parallel Execution Speed** [03:22] — Running sub-agents in parallel reduces time significantly – up to seven browser tabs are opened simultaneously, with tasks completing in under two minutes.
- **Results Under Budget** [04:45] — All items were found for under $1,300, well below the $3,000 budget, and the visualization showed the room with the new furniture.
- **CAPTCHA Solver Skill** [06:23] — A custom CAPTCHA skill was built using CDP frame introspection, high-res screenshots, and precise coordinate clicking to solve reCAPTCHA challenges (e.g., selecting traffic lights and buses).
- **Temp Mail and Reddit Account Creation** [09:52] — Using tempmail.lol, the agent creates a temporary email address and then navigates to Reddit to create an account, bypassing any email verification at signup.
- **Posting a Meme with CAPTCHA** [12:39] — The agent generates a meme using Nana Banana, uploads it to imgur, solves a Reddit CAPTCHA, and adjusts the post title after reading community rules (no AI-generated content).
- **Learning and Reusability** [15:35] — Although the Reddit posting process was slow and required iteration, the presenter emphasizes that skills built during the session can be reused, making future attempts faster.

### Conclusion

The video showcases how Claude Code can orchestrate complex browser automation workflows, from parallel product searches to CAPTCHA handling and social media account creation, with reusable skills improving efficiency over time.

## Transcript

thought again, I wanted to do a bit more of my browser automation testing with Cloud Code, the kind of the skill set up we have. So, I came up with kind of four first thing I'm going to try is kind of
a pipeline that kind of kind of involves Amazon running kind of maybe in parallel. We'll see how that works to find different items that kind of fits into my existing space, and we can pipe that with Nano Banana to kind of try to
update my already existing space. And it should find items that kind of matches, and we can get the price and everything. So, I wanted to see how that works out. kind of CAPTCHA solving, and kind of do some tool building on the fly to solve
interesting problem. Also, wanted to test out can we use temp mails to kind of create Reddit accounts on Yeah, basically see what we can do.
and we can actually create an account using temp mails. Haven't tried that with my browser automation test. And I had a look at parallel operations. Think up. So, we also going to check out parallel operations in the browser with
multiple tabs, and see if we can kind of keep Maybe we run some sub agents on should get into it. Let's start with the Amazon pipeline, and we kind of see a bit more what I mean when we get into it. Okay, so if you haven't seen any of
my browser automation videos before, basically we are just using Cloud Code with some setup we have here on You can basically read more about it, but tool with Chrome developer setup, so we can kind
of run You can see here we for example have commands here we can run using kind protocol, to actually control the browser in like this developer mode. So, this is kind of how we Yeah, you can see we are using the remote debugging setup
on Chrome to actually control the browser via JavaScript, right? you can read more about it at my Skills.md story if you want to. But basically, now what we can do is we can just use Cloud Code to kind of
use the skills to kind of navigate this browser here. So, like I said, I've been testing out the the Amazon setup, and this is kind of my base image now. Yeah, I can load that.
base image. It's just a living room. So, I want to see now if we go to Amazon, and let's say we say find some other furniture here to kind of fill up this image, and we want the price. Maybe we can set some kind of ceiling on the
price here. So, let me just come up with the first prompt here. a new chair, table, couch, and some curtains in a Scandinavian style. All items must match my current living room, and we kind of tag this image. Also,
create a visualization with Nano Banana of the items and a budget of $3,000 max. Okay, so now we kind of we kind of have this Amazon skill now, so we should just navigate over to my browser, right? And start doing some work in Amazon to kind
hopefully all of this should be automated now, and this is kind of the only input we have to do now. So, you can see we loaded our Amazon skill. Perfect. And we set some budget distributions here. And then you can see
we are doing this in parallel. We are opening maybe four browsers at once, and we are seeing all the You can see we search for Scandinavian sheer linen search for Scandinavian sheer linen curtains, Scandinavian
Yeah. couch, table, and chairs. Okay. So, now you can see we are spawning four sub agents to kind of look at each of was new. This is what I wanted to talk a
bit about parallel browser automation because now we can see we are running because now we can see we are running these sub agents here on each single this over the weekend. This kind of speeds up everything we do with
automation in the browser because we can do all of these things in parallel. That kind of saves us a lot of time. So, that is super interesting, and it's something now, let's just let this run and kind of
kind of end up with an image that kind of solves everything we wanted to do and stay inside our budget. So, you can see we browsers here, right? So, we are up to like six, seven browsers or something.
of things at the same time here. And I think this is happening because we can do all these parallel agents. Okay, so you can see now two of the agents has have finished our coffee table and our chair. So, now it's just the sofa and
the curtains left. Okay, so all the agents has finished. We closed back basically all of the tabs, and now you can see we ended up with like $1,300. That was way under budget. Okay? But Yeah, I guess that's fine. And now you
can see we're going to go into the to generate the visualization with using File AI and Nano Banana. And then we're going to kind of kind of compare the the first living room and kind of the second one when this is done now. Okay, so you
can see this was the image we had. There were no curtains. It's a couch, the look now. Okay. Yeah, this is pretty good. What about the table? Yeah, I think it's changed at least
Yeah, this is a different table. It's a different couch. You can see we now have And we have the chair we wanted. So, you can see this is working. This is pretty cool, right? So, now we can kind of do all of these automated, and we can kind
of end up with a visualization of how all of these items will kind of look in this room here. This table, is that correct? No, not really. I guess it missed kind of the table. But yeah, that was not
something, right? So, I thought this was pretty interesting, but the main point was to see all of these sub agents working in parallel. And actually doing this in parallel, saving your time instead of doing this
sequential. So, that was pretty interesting. So, yeah, pretty happy how this turned out, and I'm definitely going to explore this setup more. But now, let's do a quick test on these CAPTCHAs because that is something we
before we try this, let me just explain a bit. So, this is kind of the CAPTCHA seen this one, right? So, basically what I did yesterday, right, in the testing, um I went to Cloud, and if you look at the skills
here, you can see we have a skill called CAPTCHA. Okay? CAPTCHA. So, this is solve CAPTCHA challenges on any page using CDP frame introspection, high-res screenshot, and precise coordinate clicking. So, basically what
we did ahead and did, because remember I said if you want to build a tool around this, this is perfect. And it created a solver script that was called CAPTCHA solver. So, we have this script here, but I think it's better to actually look
in the skill here. And this is kind of it figured out how to solve this by kind we're going to see how this works. You can see it's doing like a bigger screenshot than just a screenshot of the full page. So, what I basically what I
went ahead and did is I just navigated to this Google reCAPTCHA demo here, and just let it run and run and test and test until Cloud figured out a way to do it. And from there, we just created our skill now. So, let's
just see how if it's going to Sometimes you need a bit of a testing to do this, but let's try it out now and see if we can actually solve this. So, to solve in tab in a browser. Please start working on this. Use skills and
tools. Okay. So, I want to see what kind of skills it picks now, right? You can see it loaded up the CAPTCHA skill. That's exactly what we wanted to see, right? And now it's going to go into the check out what tabs are open. You can
see it's going to click on this, right? And now, hopefully it's going to launch the tool. You can see it's doing some screenshots here to kind of see Yeah, this is up. And hopefully now it's just going to go back and pick the tool it
created yesterday to try to solve this. You can see there it kind of flashed. It kind of zoomed in on the image to get like a high-res image. And from there now, you can see we are looking for the traffic lights.
And it kind of does like an overview here, and yeah, it did select all of those. We need to do one more. This time we want to find buses. So, hopefully again, we're going to just run the tool. We're going to see like a
short flash, and we'll screenshot. Yeah. And you can see it kind of checked here. Solved false. So, it didn't pass. But now it's going to try again. So, now it's going to look for bus buses. Yeah, it finds some buses here, a couple I
think. And now it's done. So, you can see it is working. It's super fast also because this is a bit important in CAPTCHAs to try to be quite quick because you have this limited time window.
And it was super fun. I regret I didn't actually record that because yesterday when we started pounding on this, it spent a lot of time actually figuring out how to build a tool. That was a bit sad I didn't actually
record that, but you can see how well it worked in practice now. And you can also think about that this could be adapted into other different CAPTCHAs and stuff like that. So, just very interesting to see the kind of the way
it worked around this and built the tool it needed, so we can reuse that in the future. So, this is something I I was pretty pleased with to see that it can solve. So, I want to do the final thing now, and that is kind of the temp mail
Reddit account just to see how How can we push this now. Okay, so let's do this pretty simple. We have a challenge create a temp mail at tempmail.lol. Go that to create an account and create a meme with Nana banana and use the file
AI and post it on a fitting subreddit. Build and use skills and tools if needed of course. So I think this is a pretty interesting challenge because of course you need to actually create an account. And I think this is something worth
testing because this could be like a big issue, right? So now we're just going to start on tempmail.lol. And yeah, you can see you can see we have some old emails here from you my yesterday testing.
But that should not affect it, I think. You can see we get our temporary email And from there we should just navigate over to old Reddit, I think. So the more is I wanted to see if there are there any
actually like blockages or something like that to actually stop this from being automated. On these accounts, I'm not quite sure. So now you can see we are navigating over to Reddit. We should have our I
know this is in our region, but we should have our email, right? And from here now we should just be able to use like CDP, put in our email, works. So this is something I want to check out. Does Reddit kind of block
temp mails? Or can this be used by Or can this be used by these automated browser agents, right? So it did bent ahead now and actually did create a meme before we did account.
I don't know about that, but okay, fine by me. So just create like a temporary meme.png. Perfect Drake format, okay. Uh but it still needs to create account. That was kind of the main thing here. So
navigate this this page here to actually fill in like the email address and continue. Okay, so there we can see it found the email address. It put in kind of our temp mail. And now it just
needs to find a continue button and maybe that's it and it should get the email confirmation or something back on the temp mail. And now we're just going to do the username. We need a password. Okay, so
now it's going to try to fill out some kind of password here. I don't know what see if it can do this. Yeah, you can see there it kind of need to click continue here. Is there no confirmation email? No.
Not really. Just going to click on that. So that was it. That's pretty strange. Okay, so it skipped that and that's it. Now we created our account. And remember
the final thing we wanted to do was actually post the meme. That was much easier than thought. Why isn't there any Hmm. I was sure there was some kind of ish some kind of hurdle we had to get
over here to try to solve this. Not a capture anything. That's strange. Okay, so now we got a confirmation email. So we go over to confirmation email. So we go over to programmerhumor/subreddit.
in the submit thing here and there are some things we have to fill out like some things we have to fill out like URLs, image, title. AI generated this meme about AI writing code we come full circle. That is the title. Okay. Okay,
so we put in kind of the image URL. Oh, here's the capture. Okay, that's interesting. So now I want to see if we can actually pull up the capture tool we interesting. And we also need to choose a flair. Yeah, that's the flair.
So you can see can navigate these drop downs. Selected meme of course. Yeah, we got the flair. Now we just need to solve the capture. Let's use the capture skill. Yeah, that's good. I like to see that. And it clicked it and that's it.
fine. So let's see now can we actually submit this now? It's solved. I can see a green check mark on the capture. Ah, there's some issues. It needs to be like a
imgur, pastebin, stack overflow, TikTok, YouTube. Also something about the requirements for this community. It's going to be interesting to see if it tries to solve this. Maybe it goes to pastebin and does something with the
image. Let's see. So it's uploaded something to imgur here. That's pretty Yeah, we got the successful imgur up update upload. Okay, so now we have a new one. Now we're going to find cars. This should maybe be a bit better. Okay,
so that was very quick. I didn't even catch that, but it did solve it. It just took some time. Sorry about that. So now I want to see if it's just going to click submit again. So now we have the imgur URL. We have
We solved the capture. So now basically all it has to do is click submit again. So here I had to understand that there's something wrong with the title. So it just said just
read the rules and fix the title, okay? It doesn't have to change the URL because that's good. So it just needs to find the rules and then it's just going to change the title. So now it goes to rules for
Uh that's good. So hopefully it's just going to read to figure out what kind of title we can do. No AI generated content. Okay, so that's the issue with the title.
So that was it. Doesn't seem to be anything here, but let's click the post anything here, but let's click the post now. That was a bad meme though, but we did actually do it. So that was pretty cool.
Uh it took some time. I wouldn't say it was very efficient, but we got there in the end after a lot back and forward. But this is what is interesting, right? check. So
it's very slow. It has to do a lot of operations back and forward. But if I kind of went in here now and I said take all the learnings from this and turn it into a skill. So the next time we're going to do this, it's going
in previous videos and I know that works. So that is a kind of a solution. like with me, it's going to take long time, but the second time I do it if I kind of update the skills and save some memory and stuff, maybe some temp code
the next time you're going to do that like we saw with the capture tool I created. So yeah, I'm kind of happy with all the things we tried today. So yeah, like I said, if you want to check more into this, I have some more
Uh if you want to try it out for yourself. This is just a small side project I have for my agent that is trying to sell these small skills here. wanted to test out today and I really enjoyed it.
So I'm in San Francisco at the moment. So maybe the audio is a bit bad. Hope it's not too bad. I've just been to the GTC. So there is still opportunities, I
think maybe today or tomorrow to join in on my DGX Spark giveaway. So definitely look out for that. You should find the link in my previous video, I think. Or maybe a few videos back. So you can join the giveaway of the $5,000 DGX Spark. So
Other than that, have a great day. Hope this gave you some inspiration of stuff you can try with your browser AI agent. I'll see you again soon.