BM Snaps

Gallery

Burning Man really tests my mantra, “There’s no bad weather, only bad clothing.” But it’s true! For one week, it’s the most interesting art and technology playground in the world, creativity unbound. As a photographer, it’s a great place to test out new gear and processes. Here’s a few snaps from the opening days.

The Temple is really special this year, very open, and the Man concept actually came from fellow Automattician, Zander Rose.

Homework from WordCamp

AI, Open Source, WordPress

A few times during the WordCamp US keynote, I asked for a show of hands and got mostly blank stares back, especially on the AI questions. That’s not a criticism — it’s a reading list. If the future is going to be built on this stuff, the WordPress community should be fluent in it. Here’s the homework.

How models are actually trained

When I asked who knew about the reinforcement learning models go through, Robert looked out and said “I see no yeses.” Base models are trained to predict the next word across huge amounts of text; the friendly, helpful behavior comes after, in post-training, where human/AI feedback reinforces some responses over others. If you use these tools every day, understanding that pipeline changes how you prompt them and what you expect from them.

Start with Hugging Face’s Illustrating RLHF, a visual explainer. Chip Huyen’s RLHF deep dive is the best longer read, and Anthropic’s Constitutional AI paper shows where post-training went next. If you only have time for one thing, watch Andrej Karpathy’s Deep Dive into LLMs — pretraining, fine-tuning, reinforcement learning, and why models behave the way they do, all in one sitting.

Uncensored models

Almost nobody had tried an OBLITERATED model. Since refusal behavior is added in post-training, researchers found it can also be removed — it turns out to be, roughly, a single direction inside the model that you can subtract out. Maxime Labonne’s Uncensor any LLM with abliteration explains the technique, and the paper behind it, Refusal in Language Models Is Mediated by a Single Direction, is short and genuinely fascinating. Whatever you think of the applications, it’s worth understanding that open weights mean the community can modify these models, the same way the GPL means you can modify WordPress.

The open-weight wave

Only a few hands went up for Qwen 3.8. The open-weight models coming out of Chinese labs — Qwen, DeepSeek, Kimi, GLM — are at or near the frontier at a fraction of the cost, and the smaller ones run on a laptop. I demoed one from this MacBook on stage. Token costs have dropped about 20x, and that curve isn’t slowing down.

The easiest way to feel this for yourself is to run one locally tonight: install Ollama or LM Studio, pull a small Qwen model, and talk to an intelligence that lives entirely on your own machine, no cloud required. For a community built on the four freedoms, that should feel familiar — and exciting.

Read one, watch one, run one. Then bring what you learned to your next meetup.

It is really important to me that WordPress community members ride this AI wave.

Errata: I said Automattic has 32 or 33 data centers; the actual number is 29, all co-located.

Surprise DM

Automattic

When I saw the Slack notification preview, “Matt, I’m literally crying right now,” I must admit my first reaction was “Oh no, what happened…”

Matt, I’m literally crying right now. I just read the Baby & Caregiver Attendance policy that Automattic will fly your baby and whoever is caring for them to be with you at a meetup while you’re breastfeeding.

Everyone talks about closing the gap for women in the workforce after kids. This is the first time I’ve seen a company actually pay for the answer. ❤️😭

I don’t think you hear this enough, so I want to say it plainly: thank you, THANK YOU. Forever grateful.

Stephanie Ospina also posted about it on LinkedIn. This benefit isn’t even on our public benefits page! From our Field Guide post revisions I found it was added in 2018 by Ingrid Miller of our Human Wrangling team, and it’s a great example of the care and effort they put into shaping what we call the Automattic Operating System.

OmFest Registration

Personal

Thank you to everyone who expressed interest in shaping OmFest. We’re still working out some details, but we’ve got a Luma registration page for the event up. Please join on September 29, 2026, Om’s 60th birth anniversary, for an eclectic conference curated by his different communities to celebrate the many facets of Om.

If you’re just tuning in, our dearest loved one Om Malik passed away in June. This grieving process has been a rough one for me, but it has inspired some positive life changes. I’m traveling with a real camera again! A Nikon D6. I’ve shot tens of thousands of frames on the D-series, so it feels like home, but it also has some nice new features like built-in GPS (I used to use an attachment for that) and USB-C.

I’ve also become very driven to be meticulous. I miss Om’s harsh but often true feedback, but his voice in my head is a good reminder that if you do something, it’s worth doing it well. I’m kinda a hot mess (still blessed! 😂) so this will be a journey but the progress already has been rewarding. Excellence is a habit. We can always hold ourselves to a higher standard.

I have so many tools for organization and management across Asana, Google Workspace, P2, and Linear, but they’re not really connected. I’m trying to get everything in one place so there’s no data duplication, a single source of truth, and everything is always linked and accessible. It feels good, like spring cleaning, to go through everything. Woo has an initiative they call Mise en place with a similar idea. Which is why you’ve also seen those bug fixes and clean-up happening. With AI, this should be easier than ever.

Our Core Division

WordPress

Inspired by Automatticians James Kemp and Dave Smith, I’ve been thinking a lot about how much of our discord in the design of WordPress stems from the differences of our two big tribes: Bloggers and Builders.

Many of the OG developers of WordPress, including myself, tend more towards the blogging side. Writers, photographers, podcasters, journalists, the people for whom a reverse chronological stream on the homepage is the most obvious thing in the world. We link out a lot, love comments, social features and think a lot about the editor as a writing environment.

Builders are the site shapers, webmasters, the agencies, the companies, the content managers that think in a more page-first rather than post-first way. Tagging? Blogroll? Date-based permalinks? Who needs them. Give me custom fields, rich templates, a site editor, to craft beautiful sturdy pages that will stand the test of time. A store with a catalog of items where what changes is the inventory and ratings, not the content.

The Builders right now dominate our core development discourse, which makes sense. They hold a lot of the economic share with merchants and rich websites. They’re also more loyal. The Bloggers love their WordPress as the place they truly own and call home, but Bloggers have an account on every channel where there’s an audience, they syndicate out with an ephemerality mindset, because that reader/listener/follower connection is the spark that keeps their flame alive. A Builder would never be a digital sharecropper on someone else’s domain.

Dave Smith has been doing some very interesting explorations of a simpler site editor, and you hear his Builder bias around 3:20 in how obvious a static homepage as default seems to him.

The good news is we can coexist as we have at least since WordPress 1.5. We just need to give a little more love to the Blogger side, there’s so much cool stuff we could do in the Fediverse with RSS, ActivityPub, ATProto, blogrolls (link manager is still in wp-admin!), pingbacks…

I’m trying out this thing, it’s trained on my blog posts and public talks, that you can chat with as me on Telegram. Behold: AutoMattBot.

Update: This is powered by Delphi.ai and it appears very broken right now, people are sending messages and on my side it looks like it’s responding but they’re not seeing it.

I don’t usually blog all my comments other places, but while drafting a completely different blog post I came across this great post, Michel Valdrighi: The Coder Whose Code Never Went Silent.

I had to leave a comment, because Michel’s story is not well-known.

Michel Valdrighi is the GOAT. His elegant code and impeccable aesthetic paved the path for WordPress to follow, and he did it with a prescient belief in Open Source. It seems obvious now, but in 2001, software that wasn’t just free but gave you Freedom was a radical idea that made people look at you funny. He showed me that code could be poetry, that beauty in simplicity is harder to achieve than complexity.

Mike Little and I met hanging out on the b2/cafelog Sourceforge and Freenode IRC channel, both drawn in by what Michel had created. It all started on a little island called Corsica.

Bonus: After publishing, Jetpack’s Related Post feature led me to this Best Blogging Software? post where Michel commented, months before the famous Blogging Software Dilemma. See, even back then people were arguing about static vs dynamic sites.

Possible Futures of Happiness

Automattic

At Automattic, we don’t have customer service or support, we have Happiness Engineers. 🙂

There’s also a #ceo Slack channel where anyone can drop in and request a meeting, which my esteemed colleague Damianne President did, and this morning we had an interesting chat that set my mind racing about how we might engineer happiness in the future.

One overriding assumption: Personal agents will trounce company agents in most cases.

Your personal agent can — or soon will — know all your history, your calendar, your email, your preferences, and your photos, and it speaks your language. So much time is wasted in traditional support trying to see through the customer’s eyes, or authenticate them. That goes away, but watch your thinking! It’s so easy to fall into the mental model trap of legacy chatbots and data models.

As a company, your website, your WordPress, documentation, tutorials, menus, manuals need to be amazing and up to date. That’s the source of truth, and an agent will read all of it!

The interface and APIs remain because that’s useful to both human visitors and the agents.

Whenever possible, make support public. More blogs and bbPress. I actually weep a little when I think of how many beautiful words are written by our Happiness Engineers that are only ever seen once in a chat box or email thread.

My goal with our products is that support is radically available, and you almost never have to use it.

If you’re in a support role, I think you can pick a lane based on where you get energy.

If you love looking at screens, I think there’s a future where customer service wizards are watching over an Iron Man-like interface, armies of chat bots talking to people, where they can see how they’re going and intervene or nudge them as needed. There’s a ton of work in tuning agents and interactions.

If people recharge you, I actually think this will be the greatest era. All of the basic, mechanical stuff will be automated, but I think human connection will be what economists call a Veblen good, more in demand even as it is more expensive.

I think Happiness Engineering in the future will look a lot more like community building. A bot can’t host a meetup or a WordCamp.

I was the biggest Waymo fanboy, but this year with lots of trips to and from hospitals with Om I found myself craving human connection and used Uber more than ever. (It also gave me a chance to practice the one repeatable way that money can buy happiness, which is keeping a couple Benjamins in your back pocket and surprising people with them.)

Jose-Miguel x Ryo Fukui

Gallery, Jazz

I was only able to catch the tail end, but Jose-Miguel Yamal, alongside Thomas Helton and Sam Knight, did an awesome tribute celebrating the 50th anniversary of Ryo Fukui’s Scenery at Echoes in Houston.

Jose-Miguel is the pianist who played on my senior recital, and Ryo Fukui is the pianist I discovered from the visit with Hishashi at Bar Gyu in Niseko, so it was a cool confluence! I snapped a few photos with my new Nikon D6 setup.

Important Security Update

AI, WordPress

WordPress 7.0.2 went out today with two important security updates. One is a type of pre-authorization RCE we (fortunately!) have only seen a few times in WordPress’ 23-year history; the last, I believe, in the PHPMailer class five years ago.

Major kudos to Adam Kues of Searchlight Cyber for finding the batch REST API RCE, to TF1T, dtro, and haongo on the facilitated SQL injection!

Thanks to responsible disclosure, the WordPress.org Security team was able to coordinate with hosts and CDNs to mitigate the attack at the network layer. Please upgrade anyway! But it’s a huge relief to know the vast majority of WordPress sites were protected by defense-in-depth even before the updates went out.

I really appreciate how people and organizations that otherwise might not be on the best of terms come together in times like this. (Full credits in the release post.) Everyone buries the hatchet to protect as many people as possible as quickly as possible.

I’ve said it before, I’ll say it again: security is going to be a big topic this year as the technology industry digests the incredible advances in AI models. It’s a good time to review your plans and processes, sweat the details, invest in maintenance, and hug a sysadmin. 🙂