# I one-shotted a simple website

Published: 2026-07-30T00:00:00.000Z
Canonical: https://wow.pjh.is/journal/one-shot-website

There's a lot of talk about one-shot website design. But can I go from a single prompt to a website that's *good enough to ship with no revisions*?

Turns out the answer is now "yes"—at least for a simple website where the AI barely has to do any of its own writing.

For the test I decided to make a website for a recent [paper](https://nonhumanminds.org/studying-ai-welfare-empirically/) about AI welfare.

My prompt:

**Prompt**

```
I'd like to make a website for the following academic paper:
/Users/ph/Downloads/Studying-AI-Welfare-Empirically.pdf

Basically something along the lines of:
https://www.nationalsecurity.ai/

Other great examples of the kind of thing I have in mind:
ai-2027.com - overkill for our case, but still good.
situational-awareness.ai - relevant because it's another _very long_ paper
https://transformer-circuits.pub/2026/workspace/index.html

We can also take inspiration from websites that present long philosophy content in a good way, e.g.
https://www.forethought.org/research/preparing-for-the-intelligence-explosion
https://joecarlsmith.com/2024/01/02/otherness-and-control-in-the-age-of-agi/
https://www.cold-takes.com/most-important-century/

Please /own-it.
```

The [result](https://files.pjh.is/ai-wow/2026-07/saiwe/fable-v1/index.html) was good enough to ship (username "doge", password "wow").

![Landing page for Studying AI Welfare Empirically, with the paper title, authors and links to read the report or download the PDF](https://wow.pjh.is/static/images/blog/one-shot-paper-site-hero.jpg)

![The paper reader showing the introduction, a table of contents on the left and footnotes on the right](https://wow.pjh.is/static/images/blog/one-shot-paper-site-reader.jpg)

Of course, I would want to improve it. But it's solid. The fundamentals are there, including all the critical launch details, e.g. social share cards. I found no unacceptable bugs or omissions.[^1]

[^1]: There *is* one feature omission which I consider borderline unacceptable. On a phone, once you've scrolled deep into the paper, there's no button to reopen the table of contents.

To get this result, my `/own-it` skill was critical. The skill instructs the agent to anticipate the feedback that I'd otherwise have to give it along the way to task completion. In practice that means more adversarial review and testing than it'd do by default, and sometimes a kick-off interview to clarify the requirements (the agent chose to skip in this case).

I'll share the `/own-it` skill in a future post.

For now, I'm just sharing this as a notable success story. With a few hours' work on a scaffold, one could launch hundreds of great academic paper websites per day.[^2] Soon, more complex websites will be possible. And soon, everyone will have this ability.

*Note: the paper's authors didn't ask me to work on this, and search engines are blocked from indexing the output I linked to above.*

[^2]: To launch the websites, it'd be easy to make a `/ship-it` skill that'd reliably buy a domain name, set up hosting, and deploy the site with just a single prompt. That step is much easier than the above.