Highlighting 3 AI Projects Happening at Automattic

Dave:
Welcome to Do the Woo. This is an All Things WordPress and WooCommerce podcast, and this is the Emerging Tech series. I am your host, Dave Lockie, and for this episode, we’ve got Thomas Shellberg, who is representing the Applied AI team from Automattic, and therefore one of my colleagues. Hi, Thomas. Thanks for joining us today. How are you doing?

Thomas:
Hi, Dave. I’m doing great. Thanks for having me.

Dave:
Perfect. So today we wanted to talk a little bit about what Automattic is up to with AI, and you are the best person I can think of to talk us through that. So, I guess maybe first let’s just start with you. What’s your background? How did you come to be at Automattic, and when did you join the Applied AI team, and why? What were you doing before, and what was so appealing about jumping into this team?

Thomas:
Yeah, thanks. So, I’ve actually been at Automattic for nine years now. Before then, I was working for WooThemes, which was before it became WooCommerce and before it was acquired by Automattic. I’ve been working with Woo for a long time. Yeah, one of the originals— I think we were about 35 people. It was a very scrappy, small company when I joined. I was hired as a Happiness Engineer in 2014, and I worked there for five years before I started to get really interested in programming. I became essentially the first junior developer at Automattic in 2020 and became a full-time engineer. I was working on Woo Pay and some other Woo payments and projects, extensions, etc.

Pretty much after GPT-3 came out, I started to get really interested in the LLM (large language model) side of machine learning and AI. Before that, I had done some AI courses, the more traditional machine learning courses by Andrew Ng, from the famous Stanford course. But yeah, I started to run some experiments and began thinking about ways that WooCommerce could utilize this new LLM technology and some of the magic that apparently came out of this “black box.” I was quite vocal about my experiments and ideas, and we eventually created a Woo AI team—sort of an experimental, short-term team to explore what kind of projects in WooCommerce could utilize AI.

Our biggest achievement was the Woo AI plugin that we shipped to about 10,000 customers, and it still gets used to this day. The plugin mainly helps with generating product descriptions, product titles, categories, and even includes a background removal tool for images. After that, I moved into the Applied AI team, which focuses more on WordPress.com functionality and how we can improve it. I’ve been on the Applied AI team since May.

Dave:
What a lovely story! Coming from a Happiness background and ending up really in the thick of it, the cutting edge of what we’re trying to do as a business—it must feel like a great ride you’ve been on.

Thomas:
Yeah, and I think everyone should do some kind of tech support or customer support in their lifetime. It really helps you develop a level of empathy for your customers that you can’t replicate just by using the product short-term. I think it really, really helps in development.

Dave:
Yeah, it’s almost unlimited in the number of ways someone can misunderstand, misuse, or break a piece of technology. I’ve done a lot of customer support myself over the years, so, yeah, strong plus-one from me there.

Thomas:
You can never fully predict how people will use your products, especially something as flexible as WooCommerce or WordPress. People find interesting ways to use a product that you never thought of.

Dave:
Totally. My background is in agency freelancing and then running an agency, and I always used to say there are a million ways to build a WordPress site, but only a handful of ways that are actually right or good for the long term. I think that remains true to this day, and it’s definitely something I’m passionate about in my day-to-day work—figuring out when to be more opinionated about how to build successful sites and stores using Automattic’s suite of products and services.

Obviously, there are different tribes or mini-ecosystems within the greater space. For example, if you’re building on Elementor, there’s a whole bunch of stuff that comes along with that. But Automattic has its own de facto set of opinions. We think that if you’re going to do e-commerce on WordPress, you should be using WooCommerce, Gutenberg, and Jetpack. Trying to find that balance between flexibility and customizability while also having opinions and guardrails to ensure product compatibility and a consistent user experience—both on the front and back ends—is a really interesting problem to solve.

Thomas:
I think that’s a really interesting point, and it’s very relevant at this moment. We’ll get into some of the projects we’re working on, but that’s exactly what we’re dealing with—finding the balance between freedom and opinionated patterns, presets, or decisions about how a website should look and what an AI should do for you, versus leaving things open for customization.

Dave:
Awesome. That’s a nice and natural segue. Why don’t you tell us some of the things that you’re up to—maybe some of the things you’ve shipped and are iterating on, and maybe some of the active projects you’re working on now? Then we can leap into the future a little.

Thomas:
Sure. I’d say we have three major projects. One of them is public, and one is sort of behind the scenes but kind of public. The first one is the Jetpack AI Assistant. It’s a contextual AI assistant that lives inside the Gutenberg editor. It’s highly contextual, so if you select a text block, you’ll see AI options like “translate this text,” “rewrite this,” or “correct mistakes,” which is really cool. It’s built into the editor, so it doesn’t add any extra level of abstraction or anything unfamiliar. It’s just part of what you already know. So, that’s great.

It’s an extra plan inside of Jetpack. It can also generate images, helping to make your pages and posts more interesting and appealing. Instead of hunting for a stock image that might not fit perfectly, you can use AI-generated images that match your needs. It’s really good for outlining posts too. Usually, the hardest part of writing is starting from nothing, and having an outline really helps.

Dave:
Yeah, I totally agree. There’s always the challenge of a blank piece of paper. You have a rough idea, but how do you shape it quickly? When you’re writing casually, you can leave things half-finished and come back to them, but if you’re a professional marketer or under pressure to produce content regularly, you need to quickly form a good idea of what the end product will look like. I think AI helps with that.

Also, it’s great for keeping you in a particular mode—whether you’re creating or editing. I often find that when I’m in a creative mode, the editorial side of my brain starts creeping in, and it gets distracting. With an AI assistant, I know I can focus on the creative part now and rely on the machine to handle the editing later. What kind of feedback have you received about how the Jetpack AI Assistant is being used?

Thomas:
Oh, that’s a good question. I’d have to get back to you with specifics, but I do know we have quite a few users who love it. Feedback has been good. I don’t have the numbers right now, but I can get them to you.

Dave:
Cool. Sorry to put you on the spot there. No worries! Hard questions, even if we’re on the same team, Thomas. Good stuff. Alright, so project number one is the in-editor assistant. What’s project number two?

Thomas:
Project number two is called Odie. Odie is basically our in-house WordPress.com chat system. You can think of it as a chatbot platform. It has access to and is trained on all WordPress.com documentation, so it provides accurate answers to common questions from WordPress.com users.

One bot, called WaaPu, is available via a chat interface natively within the WordPress.com admin area. When you search for help, that’s the first thing that pops up. It’s a conversational interface, and it’s actually doing quite well. We make it clear that it’s powered by AI because we want to be as transparent as possible. The main benefit is that WordPress.com users get help super fast, often resolving their issues right away.

About 30–40% of support volume is handled by this chatbot, with high customer ratings. It’s really good for repetitive questions that come up from documentation and don’t require creative troubleshooting. People appreciate getting immediate answers rather than waiting for a response from a human.

Dave:
That makes total sense. If you’re in a rush, you’ll settle for a quick coffee out of a machine rather than waiting for a handcrafted brew. I guess ticket completion rates are higher than expected?

Thomas:
Yes, higher than I expected! What’s cool is that our Happiness Engineers (HEs) help improve WaaPu’s quality. Tickets answered by Odie are reviewed by HEs, who give them ratings, so it gets better based on what our Happiness Engineers would reply with. There’s a human feedback loop improving it.

Dave:
And the name Odie—wasn’t that a cartoon dog from Garfield?

Thomas:
Yes, technically it’s short for Odysseus, but yes,

there’s Odie from Garfield too.

Dave:
Maybe there’s a double meaning there. Hopefully, it’s smarter than Garfield’s Odie! It sounds like it is. I didn’t realize we were seeing such an impact. Often people don’t know what question to ask—they know what they’re trying to achieve but not the terminology to use. Chatbots can crystallize what they’re really asking. Do you think that’s fair?

Thomas:
Absolutely. There are multiple benefits. By automating simple and repetitive questions, Happiness Engineers have more time for complex, creative problem-solving, which is more stimulating and uses more of their knowledge. When I was a Happiness Engineer, I worked on the most complex tickets at WooCommerce, and it was much more engaging than handling password resets all day. So, there’s a second benefit for people craving a bit more complexity and creativity in their job.

Dave:
That’s about moving up the value chain and building stronger relationships with customers. Instead of answering repetitive queries, we can focus on novel challenges with our customers, which adds value. It ties into that earlier point about balancing customization with guidance.

So, tell us about project three.

Thomas:
Project three is ongoing and codenamed Big Sky. It’s our most ambitious project—essentially an AI assistant for building your first website on WordPress.com. We’re targeting people who recognize WordPress.com as a great place to build a site but don’t know how to design one. Being launched into a blank page can be intimidating, so as part of onboarding, this assistant helps generate an initial site layout that looks good. Then, the user can tailor it without needing to know anything about web design.

Our goal is to make the process intuitive, even for users with no prior experience with WordPress. We’re really excited about it.

Dave:
Interesting. So, this essentially teaches users by allowing them to change something already set up? They can see how blocks are composed, and by editing the blocks, they learn how the tool works. It’s easier to tweak something than to start from scratch.

Thomas:
Absolutely. It’s very hard to start from scratch. Even if you have a rough idea, it’s easier to see something similar and make it your own. It speeds up the process massively. The challenge we’re working on is ensuring users don’t feel dropped into the deep end with Gutenberg. We want to guide them through the experience without overwhelming them.

Dave:
I can see how this reduces the tension of tweaking vs. starting from scratch. It helps users stay creative and experiment, allowing them to take control in ways that feel natural.

Thomas:
Exactly. We’re targeting business owners who want to focus on their business and the creative aspect of running it. We want to lower the barrier to entry by making website building intuitive, rather than making them learn web design to run their business.

Dave:
Totally agree! This also feels essential for the future of websites. Many other digital channels, like social media, are intuitive and easy to use from the start. We need websites to have a similar experience—something that just works without needing expert advice.

Thomas:
Absolutely. Our idea is to grow with the user. Initially, the site may not need complex functionality, but as their business grows—whether that means taking payments, adding members, or shipping inventory—we’ll be there with modular solutions that meet their needs.

Dave:
That makes sense. It aligns perfectly with our mission of democratizing publishing. Thanks for sharing everything that Automattic’s up to. I’ve learned a lot today, and I’m sure our listeners have too. Will you be at WordCamp US?

Thomas:
Unfortunately, I won’t be there. I’m in Germany and have a broken foot, so flying isn’t an option.

Dave:
Oh no! Best wishes for a swift recovery! Thanks again, Thomas, and take care.