Web development has changed enormously in recent years. There was a time when creating a website meant dealing with raw HTML and CSS and obsessing over every little pixel by styling it yourself. Then tools like Wix and Squarespace came along that allowed you to create a decent looking website by simply dragging and dropping elements.
Now, we have tools that allow you to simply describe what you want, and they go ahead and build everything exactly how you describe it. All you need to do is ideate and give directions, and he takes care of the rest. I wanted to see how far I’d really come, so I grabbed Claude Code, Antigravity, and Perplexity Computer from Google and gave them the exact same job: creating a portfolio website for myself. I used the exact same prompts and instructions, and here’s how it went…
Same message and instructions.
I’ve wanted to create a portfolio website for a while, but I haven’t had the time to sit down and learn how to create one from scratch. So this seemed like the perfect opportunity to finally put all three tools to the test and also get a portfolio website as a result. Now, I didn’t really want a generic wallet. If I did, I would have simply used a template in a tool like Wix! Instead, I wanted an interactive portfolio with fluid animations, a section with all my published work so far, and a built-in AI chatbot that a reader could use to ask questions about all my work.
So to achieve this, I used the same process on all three tools. Since all the information I’d like to display on a portfolio website is already available online, I first asked each tool to find everything it could about me (and I mean everything). I told them that I am a journalist, to find my published work online and to research everything they could. Then, once they had all that context, I gave everyone the same message describing exactly what I wanted.
Do you want to stay up to date with the latest in AI? The XDA AI Insider newsletter is published weekly with deep dives, tool recommendations, and practical coverage you won’t find anywhere else on the site. Subscribe by modifying your newsletter preferences!
And then I let each tool do its thing. Please note that I am judging the results of the first version each tool produces: no edits, no follow-up prompts, no adjustments on my part. Just the first raw result.
Perplexity Computer
You did it all on the first try.
While Claude Code and Antigravity are primarily designed for coding and development related tasks, Perplexity Computer is something that is in a slightly different lane. Is more positioned as an alternative to OpenClawand the impressive thing is that it has access to multiple AI models. It relies on Anthropic’s Opus 4.6 as the core reasoning engine, but can intelligently route different subtasks to the model that best fits the job. Although I’ve been an outspoken critic of Perplexity, I’ve been using Computer quite a bit and have really been impressed by it.
The first task I asked of him was this: create a portfolio website for me. Now, from the beginning, I was impressed. As I mentioned above, the first thing I asked these tools was to find everything they could about me. Perplexity took the longest to conclude its investigation, but also returned the most detailed information, which is exactly what I was looking for. I went as far as to investigate my Instagram account, my Twitter, and even found some things I didn’t know about, like the fact that Authory had listed my account on their website! It was a little creepy how much the AI can find out about you from a single name, but honestly, for this specific use case, that’s exactly what I needed.
Once I sent him my idea, he asked me a couple of follow-up questions, including the visual vibe I wanted, the core audience, a photo of me, and if I had a website or portfolio whose vibe I loved. Then it took off and started building! He delegated the task of compiling my articles to Gemini 3 Flash, while Claude Opus 4.6 took care of the coding. I hadn’t specified a design vibe and gave the tools the freedom to decide. Perplexity Computer opted for warm cream and coral tones, had a photo of me on the cover, and a typewriter-style tagline that alternated between different phrases about me. This included technology journalist, computer science student, NotebookLM evangelist, professional charlatan (which he pulled from my Instagram)!
It then included links to my social networks, a Find My Work button, and a Featured Work section with a filterable grid of posts I’ve written so far and topics I cover. This included an article search bar where you could find an article by searching for it. The exciting part was the “Ask Mahnoor AI” section, which only knew about my articles. I tested it with a bunch of questions like “What does Mahnoor think of Perplexity?” and I mentioned that I have a lot to say about it and that I’m a fan but also honest about the shortcomings.
Now, Perplexity’s result was the only one that included all my published articles (as I had requested) and the only one with an AI chatbot that actually worked correctly. I had hardly any complaints with the result, and this is definitely the portfolio I’m considering implementing. It was functional, quite aesthetically pleasing, and most importantly, accomplished everything I asked for on the first try.
Claude Code built something impressive, but he wasn’t there yet
I expected something better
Claude Code has become one of my favorite AI toolsand my expectations about it were high. I started with the same message asking him to do some in-depth research and he found some pretty decent information. It wasn’t as detailed as I expected (since Claude is usually great at finding information), but for the sake of this experiment, I didn’t push it any further and moved on to the build message. He asked me 10 questions about the portfolio, including the design, the vibe, the AI model I wanted to use, if I was going to implement it, and more. Then it started to be built.
Of the three, Claude Code took the longest to build and got stuck searching for my items. Then I had to explicitly tell him to let it go and just build it with what I had already analyzed, since I was just wasting tokens! Claude opted for a dark-themed portfolio with purple and amber gradient accents.
The gradient bit caused it to give off the typical “vibration encoded” look. Like Perplexity Computer’s portfolio, it included an animated typewriter that ran through phrases like tech journalist, artificial intelligence explorer, Apple enthusiast, computer science student, and open source advocate. The last one is something I find particularly interesting. I’m surely an open source advocate, but other than that, I’ve only written about three articles on open source tools. So, this was clearly a direct result of the research Claude had done.
I liked the About section that Claude came up with and included my journey presented as a visual timeline. It included a post grid with cards for each outlet I wrote for, contact information, and of course, the AI chatbot I requested along with the My Work section. Now, the My Work section ended up with 80 articles as the tool was having difficulty analyzing them. That wasn’t what I really cared about here. I know if I had let him continue, he would have analyzed all 400+ articles. The interesting part was that when I used the chatbot in its current state, many of the articles it linked to simply… didn’t exist!
For example, I asked “any piece about open source” and got three results. Only one of them was an article I remember writing, and the other two were articles he had made up himself! When I clicked on them, they gave me a 404 error on the publication’s website. The same thing happened when I asked what he had written about NotebookLM. To put it in context, I’ve done over 180 articles on the tool and the AI chatbot it created gave me 7 results and 4 of them didn’t exist. So while the portfolio structure that Claude Code produced seemed the most polished, the mind-blowing articles were a deal breaker. What good is a portfolio if it is linked to a job that doesn’t even exist?
Antigravity was the weakest of the three.
It was just…disappointing.
Finally it was time to put Google Agent IDE, Antigravitythrough the same test. When it came time to find as much information as possible about me, the tool searched the web and got an answer in a matter of seconds. The information was superficial and, as with Claude, could have been better, but it was enough to work with. Once I shared my idea with the tool, a high-level plan and questions emerged about the design environment and how I would like the AI chatbot to work. I answered the questions and then gave the green light to start building. Antigravity used Gemini 3.1 Pro to build everything, and it took longer than Perplexity Computer, but finished before Claude Code.
Now, remember how I mentioned that Claude Code’s gradient design made it feel vibration-coded from the start? Antigravity’s portfolio looked exactly the same. Same dark design, same gradient feel – if you put the two next to each other, you’ll have a hard time telling which tool created which. However, this is what made me laugh. Although it took longer than Perplexity to build this, the portfolio included only seven items. Seven of the more than four hundred articles published! The AI chatbot was also a complete disappointment. One of the seven articles included an article about Perplexity, so I assumed the AI chatbot could at least answer one question about it.
So, I asked him what he thought about Perplexity and he said:
Hmmm, I couldn’t find an exact match in my portfolio for that! Try asking about ‘NotebookLM’, ‘ChatGPT’ or ‘eReaders’!
I really don’t have much more to say. The “About” section didn’t exist (although there was an option at the top, but clicking on it yielded nothing), the portfolio contained 7 articles, the chatbot didn’t work, and the layout seemed to be vibration-coded.
I didn’t expect this
What I find really ironic is that Perplexity Computer used the Gemini and Anthropic models to build their portfolio (the same models that power Claude Code and Antigravity), and still emerged victorious. It outperformed both tools using its own technology, despite the same prompts and instructions! I wouldn’t have expected that.





