Subscribe
Share
Share
Embed
The WP Minute brings you news about WordPress in under 5 minutes -- every week! Follow The WP Minute for the WordPress headlines before you get lost in the headlines. Hosted by Matt Medeiros, host of The Matt Report podcast.
Eric Karkovack (00:01)
Hi everyone, and welcome to the WP Minute. I'm Eric Karkovack. Today's episode features a segment from my interview with Chris Poteet Chris stopped by to give a live demonstration of Quill, a native WordPress app for Mac OS. The app was designed for distraction-free writing within a powerful interface. Now you can catch the entire interview over on our WP Minute Plus channel. Visit the WPMinute.com for all the details.
Chris Poteet (00:30)
The other half of this application has obviously a a broader cultural conversation, and that is the role of artificial intelligence in content creation. And as you know, WordPress is they came out with native AI connectors.
Their Jetpack plugin, which as you saw, I love and use, have their own AI credit systems that you can buy into for content generation, image generation, all those types of things. I wanted to bring some of that in here, but I wanted to do it in a way that made it tasteful and made the writing experience better. So there's three parts to it. And it is dependent right now on an anthropic API key. I actually just changed last night that came out in this release.
Where when you go in and you have the settings, you can obviously control your authentication to the browser, but you can specify an anthropic API key. And now it used to be hard-coded
Eric Karkovack (01:18)
Okay.
Chris Poteet (01:19)
to an older version of the haiku model, which is the cheapest, smallest model. so no one's running out their API keys. Now I just decided that if someone wants to run Fable or Opus or whatever they want to do, they should be able to do that. So I was waiting on the five, five haiku release, which is a lot better. So that's here now. So you can choose the model.
You can choose the reasoning level if you if that's supported in the model as well. And then another thing that was important to me is I'm going to show you some content rewriting and generation features. I don't want generic slop. Like I want the model to be able to know and how I write, how I think about publishing as best as it can and to come much closer to that. So built into Quill is the ability for you to select up to five posts and it will create for you a system prompt
that is sent in with the content generation and rewrite features to match your style as well as the evaluation I'll show off here in a second. So that was important to me to if you're gonna use AI generation or AI editing, get closer to how I write and think. That was a goal of mine.
Eric Karkovack (02:17)
That makes sense.
Chris Poteet (02:18)
yeah. So over here, you I just show these off very briefly. You could do a couple things. If you want to, you can go ahead. I'll click continue here. I'm not gonna actually do this. You can generate a whole post and you can give it, you can write a normal prompt and give it specifics about use tables, fact check. It it supports web search as well to be accurate and up to date if you wanna do that. And I know some people have different feelings about that. I don't condone anyone slopping out in their WordPress site, right? Like
To me, this is something where if you're stuck, you're looking at a blinking cursor, how do I start? This is one way to do that. The other things that I'm perhaps more proud of are the evaluation of writing quality and content rewriting. So, for example, when I click
Eric Karkovack (02:58)
Okay.
Chris Poteet (02:59)
the content evaluation button, it will send my writing style prompt plus a new system prompt that I rewrote with Claude Code and give you categorized feedback based on what you've written in several different areas. So it'll look at things like,
What are things that can be fixed? What are things and ideas to be improved? And then just last night came out with the ability to check certain facts. So you see here, it's actually intelligent enough to know that your voice is being reflected in what's being written. Like from what it understands about me, I see that here as well. And I can go into each of these areas like fixes, and I can say, hey, this is what I think you should write based on your past writing experience and also just general best practices. And I can choose.
To apply and move through each of these. And then one more.
Eric Karkovack (03:42)
So it's more like a I'm sorry, I was gonna say it's like a more
personalized version of Grammarly, it's it looks like.
Chris Poteet (03:48)
Yes, it
that's that was the inspiration, right? Without someone having to cause their API costs are kind of exorbitant and it's a whole nother tool. And you could use Grammarly if you want on the canvas, but you're absolutely right. It's a it's a catered Claude specific way. And I I specifically chose Claude because I think it's the best pros at best at writing pros of the current Frontier Labs. So that's what I wanted to do. But you're absolutely right. Like it's those same categories there as well.
And then lastly, there are a whole bunch of things that you can do inside of your post. Like let's say, for example, you want to take some section, make it longer, shorter, rephrase, fix spelling and grammar. And again, the app has a full spelling utility inside of it. You can convert. So these are all little touches that helps make AI content generation rewriting less sloppy and more catered to individual users.
Eric Karkovack (04:38)
So how long did it take you to come up with that that ability to like just grab five of my posts and catch you know, get a sense of my voice? How how how did that come about and how how hard was that to to implement?
Chris Poteet (04:51)
Well, the hard to implement was not as bad for me because obviously Claude Co. was writing it. And I should say at the outset, like the GitHub repo is public. I have a very extensive privacy policy. If someone has concerns, like it's all detailed. Someone you can turn your own agent at the repo, get its evaluation. Like I want someone to feel comfortable that they're writing in a safe and secure way. But the idea was that I wanted these things in there, but I I I wanted a voice reflection. And a lot of the tools.
Like the first-party tools from like OpenAI or Claude, they don't learn your voice as well as I wanted. So I had to do a lot of work with Claude Code and the API and specifically adjusting those system prompts to take out the categories that I felt were most important to generate a writing voice. And then the 2.3 version, I've completely rewritten with Claude over a lot of hours all those system prompts. And by those I mean they're sent with whatever you send. So if you write a make longer
action, it takes that along with what it's the model is telling it to do and gets a result. So the answer is that the idea lived in my head, but it took a lot of time working with Claude to really get it to where I was happy with it. So I I hope that's a unique contribution of the app.