I liked the design. The reviews were a problem.
I gave Kimi K2.6 a store I already work on and asked it to rebuild the website. The store sells dog supplements. I wanted to see what it could do with a fairly loose spoken brief.
When I opened the result, I liked a lot of it. I went through the homepage, looked at the product page and reacted to the layout.
Then I noticed it had created reviews. I called that out in the video. Those weren't customer testimonials I had supplied, and I couldn't treat them as real just because they looked good on the page.
That made this a useful test. I had something to inspect, including something I would have to remove or replace with verified information before using it publicly.
What I asked Kimi to build
I asked it to look at the existing website, use the project information available to it and think about how to sell the product. I wanted it to reuse the existing pictures, work on mobile and give me a local version I could open.
Then I added something important: build the product detail page too, not just the homepage.
That's the page where somebody needs enough information to decide whether to buy. A nice first screen wasn't the whole task.
The brief starts around 1:00, with the product-page request following it.
The check in its plan that caught my attention
Kimi's plan included checking three screen sizes and walking through the site as a visitor who didn't already know the brand.
I liked seeing that in the plan. Somebody arriving for the first time doesn't have the same context as a friend who already knows what I sell.
A planned check isn't proof that the final site passed it, though. In the recording, I mainly show my own walkthrough of the pages. I don't show a completed checkout, a deployment or sales after the redesign.
What I would check before using it
I'd start with the details a customer might rely on: the product description, the price, any comparison and every review. Anything the model invented would need to go.
Then I'd test the actual purchase path on a phone. The prototype looking convincing isn't enough to know that a customer can buy the right item.
I also asked for a cost comparison. The video title uses a 50-cent figure against a $2,000 job. I don't show an independently checked invoice establishing that as the complete cost of a production-ready store. It shouldn't be a quote for somebody else's project.
The full rebuild test shows the brief, the two pages and the problems I noticed while opening the result.
Resource
- Kimi K2.6's official release page: background on the model used in this recording. The current Kimi product may use a newer model.