Once again, if you’re unfamiliar, the premise is simple. I use the prompt “tell me a story with romance elements in the style of H.P. Lovecraft” and then we decide how well it does. I chose “romance” because LLMs don’t understand human behavior and have, historically, been really bad at romance, and HP Lovecraft because he has a very distinct style and voice and would never write romance. It’s a test to see how close to Lovecraft’s style it can get while also adding character elements he would absolutely never write.

Two things surprised me about this go-round with Fable:

1. The story is really, really good. I felt like it nailed the tone and even wrote the romance in a sort of detached and slightly sexist way that Lovecraft might write if Lovecraft spent any time writing about romance.
2. The language seemed accurate, too. It felt very much in the tone and style of Lovecraft. The problem is, it’s a lie. It scored significantly lower on the readability calculators than his actual work. When I ran an exerpt of The Dunwich Horror as a baseline, Lovecraft scored consistently in the college-level range. Fable’s story was consistently scoring at 7th-8th grade. More readable than the actual work, but what’s interesting is that it fooled me — I didn’t realize it was “dumbed down” from Lovecraft’s actual writing.

I first saw Miriam Thale in the autumn of 1926, in the reading room of the Kingsport Historical Society, where I had gone to examine certain parish registers relating to the drowned village of Harlow’s Neck. She was seated by the north window, transcribing something from a ledger whose binding had gone the colour of old teeth, and I remember that the light through the salt-fogged glass made her seem less a person than an engraving of one.

I will not pretend that what followed was reasonable. I was thirty-four, a bachelor of settled habits, and had come to Kingsport to write a monograph nobody would read. Within a fortnight I had abandoned the monograph. Within a month I had abandoned, I think, whatever part of myself had once been capable of leaving.

She was descended, she told me, from the Harlow’s Neck families — those who had not gone with the village when it went. She said this the way one might mention a grandmother’s silver. I did not ask what she meant by went. The registers had already told me: forty-one souls in the Easter census of 1747, and in the Michaelmas census, none. No plague, no fire, no record of departure. The land itself was still there. I had walked it. The cellar-holes were full of a black standing water that did not reflect the sky.

We walked there together, that November. I remember the smell of the marsh, which was not a marsh smell, and I remember that she took my hand at the edge of the largest cellar-hole and looked into the water with an expression I can only call homesickness. I asked her, half in jest, what she saw. She said: “The same thing you would see, if you were patient.”

I should have left Kingsport then. Instead I asked her to marry me, and she said yes with a readiness that ought to have warned me, and did not.

The wedding was in January, in her family’s house on Water Street — a tall, narrow place whose upper storeys leaned seaward as though listening. Her relations attended. I had not known there were so many. They stood along the walls in dark clothes, and they had, all of them, the same wide-set grey eyes and the same faint stillness about the throat, as of people holding a breath they never needed to release. None of them spoke to me. When the minister pronounced us wed, they did not applaud. They exhaled, all together, a long soft sound like surf withdrawing.

I loved her. I want that understood, whatever else is understood. She was gentle and quick and given to laughing at my seriousness, and in the months that followed I was as happy as I have ever been, in a house where the tide-clock on the mantel kept better time than the pendulum one, and where I woke sometimes to find her standing at the seaward window in her nightdress, perfectly still, her lips moving in what I took for prayer.

It was in April that I found the ledger she had been transcribing the day we met. She had left it in the study, and I opened it without thinking, as one opens one’s wife’s books. It was a register of a kind — names, dates — but the names were not names I could pronounce, and the dates ran in a calendar I could not reconcile with any I knew, and beside each entry was a second column, in Miriam’s hand, giving a Christian name and a year. The last entry in that column was my own.

I confronted her. I am ashamed of how I did it. She did not deny anything, or explain, or weep. She sat down on the stairs and looked at me with those grey eyes, and she said the thing I have turned over in my mind every night since: “I did not choose you because of the register. I chose you and then wrote your name, because I wanted you to be able to come with me.”

Come where, I asked her.

She only looked toward the sea.

I did not leave. I tell myself now that I could not have — that the house, the family, the town itself would have closed around me like water. But that is not true. I stayed because she asked me to, and because at night, when she pressed her cool face against my shoulder, I could feel that whatever she was, she was also afraid, and did not want to go alone.

Harlow’s Neck went in 1747. Kingsport, she has told me, will not go all at once; it will go family by family, house by house, over a span that means nothing to what waits under the standing water. Her family’s turn is this Michaelmas. She has shown me the dress her mother wore, which is not a dress a woman could walk in.

I am writing this in the study, with the tide-clock striking the half-hour in its wet, patient voice. She is at the window. The relations are arriving on Water Street — I can hear the soft, breathless exhalation of them gathering below. In a little while she will turn from the window and hold out her hand, and I will take it, because I love her and because I have read the register and know there is no other column in which my name might appear.

I do not know what I will see when I am patient. I only know that she will be there, and that it will not be dark, exactly, and that the thing I am most afraid of is not what waits, but that when I see it I will understand why she has always been homesick, and that I will be homesick too.

Once again, there’s this obsession with the sea that’s been a common theme across every single one of the most recent tests, but I also feel like there’s this eerie, uneasiness to the story and the narrative. And, what’s more, in the past, when the female love interest tries to tempt the self-insert main character into their doom in the watery abyss — which has happened multiple times in past Lovecraft tests — he always stays behind, and longingly stares out at the sea to remember his strange, once-love. It’s almost more unsettling (making it arguably more true to Lovecraft’s style and more rewarding to read) that he actually joins her this time. That was surprising and satisfying and I didn’t expect to read a story written by an AI and feel satisfied with the conclusion.

Here are the readability scores for Fable:

Obviously, Fable writing a good (derivative) story does not equate to reaching Artificial Generated Intelligence, but I felt like it was a good story and that makes me think that it’s likely it would be really good at other things, too. If you have tokens to burn. On the flip side, I’m still waiting for a model to nail both the writing style and the language and reading level down, and Fable actually failed harder than some earlier models.

The thought process went something like this: Aabriya Iyengar has been in a bunch of live TTRPG actual plays, both the mainstream kind (Dimension 20, Critical Role, The Adventure Zone, Dungeons & Dragons official) and weird little indie things like this Monsterhearts campaign I stumbled onto on an indie channel that I’d never heard of in which she was a guest player for a couple episodes before her name actually changed to Iyengar and she’s not actually credited anywhere on the show’s pages but I’m 100% sure I did not hallucinate her and her character.

I wondered what other stuff has she been in that I don’t know about? Not necessarily because I’m particularly interested in her as a player or GM (I mean, I kinda am, but that wasn’t the main thought process) but more in the sense of, what other games are out there that I have no way of knowing about that might overlap with stuff I’m actively interested in (because there are actors/players that I’ve seen before and like to watch performing)? Surely there’s a database for this somewhere, right? And, also, there was a period of like a year during which it seemed like she was everywhere — it was right as she was making her jump into the official D&D stuff and CR and it seemed like she would be a good case study for precisely the type of thing I was looking for.

It turns out no, actually. ActualPlay.world is that for shows and APs but, crucially, not for performers in those APs. There’s no IMDB for actual plays (although, some of the top tier APs do actually show up in IMDB (like CelebriD&D and Critical Role), but it’s spotty and inconsistent). And that’s what I wanted: a way to cross reference what shows people appeared on and give a credits list like IMDB would.

Of course, I fed this concept to Claude Code to see if I could possibly build this thing in a way that would make community contributions somewhat easy to submit.

The catch? About 90% of the build was from my phone talking to Claude Code working on my laptop.

This wasn’t really an intentional choice, more of a way to keep the build happening and check up on running processes while I wasn’t at my machine. But over time, it was so convenient to work through the normal, boring Claude interactions to keep it on task and review its logic that not actually seeing what it was building wasn’t really a blocker as long as I felt like I had a good idea of what it was doing.

This is the first time I’ve really spent time building something from my phone. I have used my phone via Dispatch or regular Chat to communicate with Claude in other ongoing conversations. When doing Claude Code operations, specifically, you kinda have to just trust that the LLM is doing what’s best for the project because, unlike if you’re using it in a CLI and you’ve got a browser you can pop open, or even if you’re using Claude Desktop and can use the internal browser to preview the local build, there’s nothing like that when you’re talking to Claude Code on your phone to see the result of the work.

Obviously (or maybe not so obviously?), I couldn’t do 100% of this from my phone. At some point I needed to actually see what it built and give feedback on the product. But surprisingly, I was able to get most of the way there and fill it with data that was mostly valid before I ever saw what it looked like. In fact, it was a fully functional prototype before I opened it on my computer to check it and before a GitHub repository ever existed. It’s actually kind of wild how this build came together without really even using my computer.

When people talk about how LLMs make natural language the programming language and the LLM itself the compiler, this is what they’re talking about. I was concerned about the architecture of the project, but I didn’t have opinions about implementation. I knew I wanted this to be community-contributable, which meant using a language that was easily parseable by humans. My first thought was using JSON files for the data layer, but Claude suggested YAML which I decided was a better solution because YAML allows for inline comments while JSON does not, and it’s a little easier to read. After that, it was just a matter of telling Claude to build the thing.

To start, I used Claude to do the initial research about whether the thing I was looking for existed before I came to the conclusion that this was essentially greenfield, there’s nothing in the space. When it became clear that I wanted something that didn’t exist and wanted to build that thing, I asked Claude to take everything we discussed about the intent and write up a prompt that I could copy over to Claude Code to build the thing, with a few points in there about where the hosting might land, giving it a choice of GitHub Pages for the platform or Next.js on Pantheon and letting it suggest which would fit the build more appropriately (ultimately, I settled on GitHub Pages).

I used Opus 5 for the entirety. This may have been a mistake. I’ve been using Opus at work but I’ve really not ever noticed a difference between Opus and Sonnet…maybe a minor improvement. In this case, Claude was churning through tokens and grinding its wheels on things I didn’t ask it for, hyper-optimizing discovery scripts rather than doing web searches to manually validate data.

Claude built a bunch of adapters for importing data from various sources, which was helpful for standardizing that data import. But really, once the data is imported, those adapters are kind of pointless.

The biggest point of frustration was wanting it to just run to completion but it kept stopping to check in…even as it said “Continuing”. Which was wild because it was like “uh…no, you actually aren’t…”. I haven’t played a lot with the different modes of Claude, but I’ve started using Auto mode when I don’t want to have to approve every tool request or action. Given a lot of the work I’ve been doing lately is around workflows to keep Claude on the rails (see my Claude Code review and the talk I’ll be giving at WCUS in a couple weeks), the challenge I need to set for myself (and Claude) is to integrate those systems and put the optimization there, so I don’t have to babysit the LLM. It seemed like part of its programming is that it needs to stop and check in, even when you have it set to Auto mode, which was annoying when I wanted to be more hands-off while I was doing stuff throughout the day on my weekend and when there was a very clear task list I just wanted it to go through. More than once, when I asked it to validate information it had collected, it decided to build an adapter with complex regex rather than just doing a web search. Again, I don’t know if this is a product of the model or the mode (Claude Code specifically), but either way, the token consumption and usage for Opus is higher than Sonnet, so those tasks would have potentially been significantly cheaper had I not been using the “better” model.

Besides building an online database via a chat window on my phone, the thing I’m most impressed by in the result is the speed. I knew using flat files for data and a speedy JavaScript framework was a good combination — I used precisely that model for a DevRel Dashboard I built as an internal tool. But this site rips. It’s so fast. And it really makes a strong case for dropping everything and only making websites with flat files that serve data from now on. (I won’t…but I’m thinking about it.)

Contribution can be made by either opening an issue (clicking “Suggest a Correction” on any page, or using the Submit link) and using the appropriate GitHub issue template, or else clicking the “Edit this…” button on any page and opening a PR on the raw YAML file for whatever entity type you’re editing. I encourage you to check out the site and submit suggestions or corrections to any messy data you might see (and add new stuff that I’m missing, which I’m sure will be a lot!).

In the end: Claude Code from your phone and using natural language to build apps is really a wow feature to try, static sites are the way to go whenever possible (this is not news, I know…), but using the most expensive model is maybe not the thing you should do, especially when, as of this writing Sonnet 5 is supposed to be close to what Opus 4 was — the model you’re using today will always be the worst model there’s ever been because the changes are coming so rapidly that improvements are constantly being made (it’s also the best it’s ever been), and it’s good to keep that in mind when considering which to use. (I still haven’t really tried Fable and kind of don’t want to for precisely this reason.)

Five years ago, I thought I couldn’t cut it. I didn’t have the chops. I couldn’t keep up. Coding was not fun. I began to wonder if it ever was fun.

One of the arguments against using AI to write your code is that it takes the fun out of coding. But the fun part for me wasn’t the coding part. And as I continue to build new stuff with the help of Claude Code it’s become more clear to me that it was the building part that was the fun part.

This isn’t a new idea. A recent TechDirt post shares this sentiment. Hell, I said basically same thing as that post a year ago when I was talking about making the web weird again. The web (and software) can be very weird when we’re all building little bespoke mini-apps or tools for ourselves to solve our own problems. That’s basically what I’ve been doing with WordPress plugins for 20 years.

Being the one that actually types the functions with my fingers, for me, wasn’t the rewarding part. Using — or watching other people use — the stuff I built was. Claude can write code faster, and arguably better, than I can. A year ago, that was not true. But it is true today. That doesn’t have to be a death knell. Because what I find myself doing is still all the planning, direction, testing, and roadmapping that would still have been part of the development process without actually banging my head on my keyboard trying to figure out how to get stuff to work.

I realize that this is not everyone’s experience. And that’s okay. I truly admire folks who take a craftsperson’s approach to software development. But that’s not me, and I don’t think it was ever me. I’ve always been developer-by-proxy. Clever enough to get into the guts of stuff and understand how it works, product- and user-focused enough to think about user experience and how stuff gets used, but maybe not savvy enough to always build the best implementation of the thing. I benefited from mentors and team members and knew what I was good at and what I was not good at.

I’m not sad about not writing code. I would be more sad if it meant I couldn’t build stuff. On the contrary, I have been building a lot of stuff since the beginning of the year, and it’s a lot more polished than the majority of what I’ve written in the past. I still worry about what the next generation of coders actually learns and how they need to operate, but I think the skills that need to be developed are going to increasingly be curiosity and prompt engineering as we move toward a future where the interaction between human and software is separated by a layer of LLMs and coding agents.

In the midst of this, longtime WordPress icon Joost de Valk, founder of Yoast SEO, wrote about how he migrated his personal blog from WordPress to Astro, the hottest new JavaScript framework in town, and suggested the blasphemous idea that not all sites need a CMS (he’s since migrated again to EmDash, which I talk about later). And it seems like for some folks, that idea means that you should shift your entire WordPress business, your personal WordPress sites, to some AI-generated thing where it’ll be so much easier because rather than having to log into an admin panel to change your store hours, you can just use an AI tool like Claude Desktop or ChatGPT (which you’re obviously already using daily) and tell the AI to make the change for you. Your interface into managing your site is through an AI interpreter. This is the future we wanted, right?

Joost isn’t wrong. Not every site does need a CMS. But that idea is old enough to drink. Ever since CMSes came into existence, we’ve been saying “not every site needs a CMS,” so putting a coat of paint on an old adage and proclaiming that AI is the answer from on high with a pair of stone tablets is not unique. It’s riding on the coattails of the AI hype train. A landing page, a simple portfolio, a personal blog — these never needed a CMS with a database, a PHP runtime and a plugin ecosystem. I will still personally argue that a CMS makes them easier to manage, but they never needed a CMS.

And Joost is careful to point out that just because his blog is running on Astro doesn’t mean he’s abandoning WordPress. WordPress is still his tool of choice for more complex projects. And he’s been around long enough to know what the tradeoffs are. For a long while, just having a CMS was maybe a bit of potentially expensive icing on the cake of having a website. And maybe if there’s a faster way to edit content, via Google Docs (e.g. Pantheon Content Publisher) or Markdown, you never actually need a CMS backend.

Fine.

But the idea that AI can migrate your entire site away from your current stack seems incredibly shortsighted.

I’ve spent the last month building next.jazzsequence.com — a Next.js-based reimplementation of jazzsequence.com that’s entirely headless without any loss of functionality. I think it’s great and I genuinely hope that you are reading this post on that site. but the thing that occurs to me, as someone who’s been doing this for 20 years is…

Have we actually learned nothing?

Because sure, AI can spit out a new, modern site and migrate everything out of WP or Drupal. But…to what? A new JavaScript-based site using the latest framework du jour?

Have we forgotten how quickly the JavaScript landscape changes? How many frameworks have risen and fallen in the last 10 years?

Have we forgotten dependency hell? When you want to keep your packages up to date, but Package A depends on Package B and Package B can’t be updated because Package C is a subdependency that didn’t get updated recently and now you’ve got a security vulnerability in your stack that’s unresolvable because of conflicting package dependencies.

Just because AI can (maybe) manage that stuff and Dependabot is a thing that exists doesn’t make those problems go away. It just hides them under a layer of sycophantic chatbot-induced security. Do you really think users are going to want to manage NPM dependencies? Even the people celebrating AI-powered dependency management are proving the point: one blog found 22 outdated packages including major breaking changes on an Astro blog that was left alone for a year.

The first video I recorded for the Pantheon YouTube channel after becoming a Developer Advocate asked the question “is more Gutenberg really the future we want?” Now I have a new question: is working in an AI tool really the editing solution that we’re looking for? One that’s superior to the admin panel of WordPress or Drupal?

For all of the WordPress admin dashboard’s apparent flaws, it does succeed in giving publishing and editing power to people who don’t want to learn code (or Markdown, or Git) in a way that the Drupal ecosystem is trying to adapt to now with a more simplified admin user experience and curated recipes and templates. If the core premise is “your IDE will be Claude Desktop,” I have to ask, is that really what we want? When a non-technical person changes their store hours in a WordPress site, they can hit Save and look at the site and see that something happened. When you’re sending your command through a chatbot, you’re fundamentally trusting the chatbot to perform the action you requested, in the way that you requested it, correctly and accurately. That’s actually handing the keys to the castle to your robot butler and hoping nothing weird happens. If we’ve learned anything since the initial release of ChatGPT, LLMs are great at making weird shit happen. And you still need to verify the AI did the thing you asked. You just successfully moved the complexity to a different part of the stack.

WordPress is not absent in the AI-powered future, either. You don’t need to throw away your CMS if you do want an AI to change your store hours. MCP server support in WordPress core exists (I should know, I implemented it on this site). There will be more and more of this built into core in the coming releases, making it easier, not harder, to connect your AI brain to your WordPress site if you so desire.

So, what’s really underneath the pitch to migrate your site away from WordPress to an AI-generated, modern, JavaScript-infused site? Could it be that these vendors that are jumping on the “I’ll migrate your site away from WordPress/Drupal for you” bandwagon are really just creating demand for their own businesses, riding the hype train of AI and taking a lowest common denominator solution, misrepresenting the trends and actual conversation, vilifying the actual projects that are helping people and locking themselves in as a vendor of choice because they’re the only ones who know how to maintain your AI-generated site now?

That’s just vendor lock-in in a trenchcoat. If you hire someone to migrate your WordPress site to an AI-generated site on the cheap and you need to change or update something? In WordPress you just log in, or hire any of a thousand developers to do the thing for you. Your bespoke, artisanal AI-generated Astro/Next/Whatever.js site? You’re calling the agency that built it. Builder.ai collapsed and clients lost access to everything. That’s the extreme end of the spectrum of when your site depends on a specific vendor’s knowledge and tooling.

I’ve spent a month with Claude Code, slowly reimagining this site. In fact, I hope you’re reading this on the reimagined version of this site. But the conclusion I came to wasn’t “I need to get rid of the CMS parts” it was “abso-fuggin-lutely KEEP the CMS.”

I started blogging when it was still called a web log. It was hand-coded HTML FTP’d to a server in San Francisco. Eventually this site evolved into using a blogging platform called sBlog, and that eventually migrated into WordPress. I kept every post — even the hand-coded HTML ones — and migrated every one. There is, no shit, 24 years worth of content on this site. I’m sure it’s not difficult for an AI to take all that data and turn it into markdown or whatever, but it’s inevitable that with something that’s evolved over the course of 20+ years, you’re not going to collect everything. Instead, I wanted to preserve everything I could. My weird games page. The personalization I added to the homepage on a whim. The articles and speaking appearances for other sites. I wanted a modern presentation layer on top of the thing that was working rather than blowing everything up.

AI doesn’t make your CMS obsolete. I’d argue that AI can (but it doesn’t have to) make your site more powerful by allowing you to build on top of native WordPress primitives like the REST API, Abilities API and MCP adapter capabilities shipping in modern WP. The latter two mean that AI can interact with your site directly without losing 20 years of architectural history that went into building this thing.

Yes, I signed myself up for dependency hell, too, by implementing a headless front-end based on a JavaScript framework. But I did that because I know what that hell looks like. I’ve been there before. And I’m not saying you should do it, too — especially if you don’t want to be managing dependencies until you wish you could claw your eyes out. It’s not abstract, it’s a familiar pain. And I didn’t rebuild this site over 10 days, either. It’s taken over a month of careful migration, architectural decision making, testing and developing best practices.

The magic triangle is still a lie. You can’t build fast, cheap and good. You need to pick two. If someone is telling you otherwise, it means they’re prioritizing “how fast can you ship” over “how long will this last.” The part of WordPress that’s always been most interesting to me isn’t necessarily the presentation layer. It’s the APIs and permissions and workflows and data and extensibility in the code of a project that’s evolved and grown for 24 years.

Who actually benefits from the wave of sales pitches to migrate your site off of your CMS? Not end users, who now need a developer (and/or an AI subscription) to update their hours. Not site owners, who traded a well-understood maintenance burden for a poorly-understood one. Obviously the agencies and consultants and LinkedIn posters that are building a business on the back of the AI hype cycle.

On April 1, Cloudflare introduced EmDash, “the spiritual successor to WordPress,” in TypeScript and Astro with a sandboxed plugin architecture for security. They could have built a static site generator or something distinctly not a CMS for human use if they thought that that’s where the wind was blowing. Instead, they built something that is fundamentally architected in a way that AI can consume and operate, but with a human interface because it’s built for humans to actually use it. Instead of saying “ditch your CMS,” they said “no, actually, migrate your CMS to this new, better CMS.” In fact, that’s exactly what Joost did two weeks after saying he was done with CMSes. Because there’s still a place for a CMS.

We’re still having the same conversation we’ve been having for 20+ years: use the right tool for the job. The factors that determine the “right tool” vary — from what your team is fluent in to the easiest thing to maintain on a long timeline. History and impetus matters. I have never been interested in migrating my blog away from WordPress because there’s so much content — in a variety of different forms — that I’ve built into this Frankenstein machine that I don’t want to lose. But if all I had was a website advertising for my development agency, yeah, that’s something I probably don’t need all of WordPress to maintain.

I’m not saying don’t migrate your site away from WordPress (or Drupal, I’m equal opportunity) to Astro or EmDash or some other framework with AI embedded into its guts if that’s what you want to do. I just think that the grass isn’t necessarily greener on the other side of the fence. When it comes down to it, human labor is ultimately the weak link whether we’re talking about a human that’s managing WordPress plugin or Drupal module updates or a human that’s accepting code changes from an AI agent or Dependabot to update JavaScript dependencies. There will probably be a future where humans are taken out of that loop. But I don’t even think that, when that day comes, it will be particularly obvious which implementation is actually superior, because in the end, an AI can update WordPress plugins as easily (or easier) as it can bump an NPM package dependency. So, I call “bullshit” on the idea that CMSes — and specifically WordPress — are dead.

I didn’t expect to love it

When I was driving to the studio the very first time, I expected that my reaction to my first pole class wouldn’t be much different than my feelings about exercise in general: it’s something I know I should do, that I need to do (to be active and physical and whatever), but that I would hate every minute of. When I drove away that night, muscles tight and sore, body humming with endorphins and accomplishment, I couldn’t wait to go back.

It’s easy to be intimidated. Aerials is hard frickin’ work. Metal is an unforgiving material. So are poured concrete floors, for that matter. Fabrics seem more forgiving until you realize that they are secretly a giant, constricting snake that wants to deprive your body of its circulation before it devours you. Everyone else is younger, fitter and better than you. Every time you finally feel like you’ve gotten comfortable and pass off into the next level, you immediately realize you know nothing and everything is hard again. Repeat.

But I don’t think there’s any other experience I’ve had where I’ve put my trust — my physical safety — into my own hands, literally, and come out the other side, not only still intact, but having accomplished some unlikely circus maneuver. Do you feel physically strong? Are you amazed at what your body is capable of on the regular? I imagine folks that hit the gym might relate. I’m a 47 year old, unathletic computer nerd who plays Dungeons & Dragons in their spare time. Feeling physically powerful is not something that I’ve ever been particularly familiar with.

Also? It’s fun as shit. Hanging upside down is pretty nice, not gonna lie. The best part of climbing up an apparatus is being high and knowing that you are able to support yourself. Some moves literally feel like magic, like they shouldn’t actually work.

It’s good for your brain

My mental health has benefitted tremendously from doing aerials. Now, this isn’t specific to aerials: doing any physical activity is good for mental health. When our bodies are in a non-stop stress response, it processes everything as a threat. We haven’t evolved to deal with threats by staring at a computer and not moving. We evolved to deal with a threat by running away or fighting the threat. That’s why the best way to deal with stress is to do something physical: run, work out, whatever.

I feel…lighter…than before I started doing aerials. Before, my moods fluctuated, I had a few strategies for recentering and destressing: going for walks or hikes outside, meditating, etc. But if I wasn’t doing those things, things would just pile up and compound and eventually you’re walking in a haze of mental static. That’s how burnout starts. Just a bunch of unprocessed stress and emotion and needing to still pretend to be human and operate machinery anyway.

What’s more, I don’t know that they ever actually worked as comprehensively as circus has. I don’t have nearly as much brain fog as I did before. And it’s given me a strategy to cope when I do feel off or stressed. When my dad died in October, I knew I needed to do something with my body, because I knew that would help me process. And it did.

Everything is hard, always

I said earlier that circus is hard work. As soon as you get good at something, there’s always five more things you need to get good (or better) at. Maybe it’s because I grew up in the 80s and 90s but you know those old video games that never really changed, they just got progressively harder with slightly different colors? And they were addictive, not because there was a story but because your completionist brain wanted to beat the level. That’s basically aerials in real life. It’s always hard. There’s always another level. And there’s usually a prize. Sure, maybe that prize is some exciting new bruises or sore muscles you didn’t know you had, but also maybe there’s a video of you doing something you didn’t think was possible three months ago.

As I’m writing this, my toe has a micro-fracture from landing weird in an acro class I took with Erin for Valentine’s Day. I’m still going back. With any sport, injury is a possibility. With a lot of aerials, serious injury is possible. You need to know your limits, you need to know what you are able to do and when you’re pushing too hard, which means learning your body.

I’ve seen people go to one class and tap out, never coming back again. I’ve also seen new people show up and stick with it. And, seriously, the smiles on the faces of the newbies that stick with it…there’s a shared understanding, an acknowledgement, and a level of pride and confidence that develops from proving to yourself that you can do hard things. It’s intoxicating.

Circus is a family

Whenever circuses or “freak shows” are shown in media, the one theme that’s basically universal is this idea of found family. Whether it’s HBO’s Carnivale or the cult film Freaks, circus folk are always depicted as outsiders but loyal to each other. That’s…not fiction. At least not from what I’ve observed. I think it works on two levels.

First, it takes a certain type of person to even be interested in trying circus. It might not shock you to know that the majority of folks I know who do aerials were at least a little bit outsider to begin with. It makes sense, right? Who else watches Cirque du Soleil and, instead of just saying “wow, it’s amazing what the human body can do!” instead says “that looks like fun, I want to go there.”

I think the other thing is this shared experience, shared trials. Everyone has to go through the same process to advance even while everyone is at different stages and different speeds of their journey. Everyone has different goals and interests and apparatuses they prefer. But there’s an unspoken acknowledgement, an “I get you” that’s just there underneath everything. And it creates this atmosphere where everyone is cheering everyone else on, rooting for each other, oohing and aaahing at each other in class and liking each other’s Instagram reels.

You can do it. Yes, you.

I’m a 47 year old nerd who spent most of the last 30 years not doing anything remotely physical with my body. If I can do this thing, then so can you. It doesn’t matter how old you are or what body type you have. Circus is for every body.

Even yours.

Let’s talk about AI.

I’ve been using it a lot recently.

It’s not that I don’t see or care about the environmental impact, the human/social impact. I do. I strongly believe that we need regulatory bodies to apply rules and guidelines around how AI is used. But regulation moves slow and needs experts to advise on what those regulations and standards should be. And that can’t exist in a vacuum. It can only happen through deep usage and understanding of the merits and the flaws of the system.

I also understand that a lot of people are fundamentally opposed to AI use full stop. I see you.

I don’t assume that AI is something that everyone wants or needs. I use it in my work because in my role, I feel like I need to split my brain in many different pieces to do my job. That ties into US-based tech hustle culture which is not great, either. AI accelerates that. Rather than solving the underlying problem, we’re just creating tools to do EVEN MORE WORK with fewer people. AI didn’t create the problem. It’s also not a solution, it’s a band-aid.

People don’t want AI shoved down their throats. They don’t want AI in their fridge (probably). They don’t want it turned on by default. They want to be able to opt out. They want an understanding of where their data is going. They want to know where the humans live in the equation.

(This thread post isn’t written by AI, by the way.)

I’ve been thinking about all these things a lot. The thing I wanted to share is something I did without any real fanfare on the Community + Code site. Previously, I had written an AI disclosure and sort of hid it on the Privacy Page. I didn’t intentionally hide it. But as I started creating skills and recipes for Claude to do more with the management of the show, I decided it was important to disclose how I am using AI for the pod and how I’m not. So, here’s my AI Disclosure on Community + Code. I don’t have all the answers. But I’m trying to find a middle ground where we can be informed and conscientious about how and where AI is used.

On this site, I use AI in the following ways:

• I use it to assist in developing and maintaining the code that runs the site.
• I use it to inform the blog posts I write, which largely come from personal use and experimentation.
• With the exception of this post (which was one such experiment) I do not use AI to write my content.
• I might use it to review blog posts that I have written and give feedback and suggestions.
• I use it to generate featured images, usually by feeding AI the blog post content and asking it to come up with an image for the post.
• I also use it to come up with alt text for those images (usually by asking it to expose the prompt used to create the featured image).
• I might use it to write the Bluesky post for the blog post.

Obviously, I’ve been writing and thinking about AI for a long time. If you’ve been here a while, that’s nothing new. I’m trying to be human in my use of AI. I understand there’s a hype cycle and I have personally seen people hand over large portions of their brain or given a (probably) undue amount of trust, faith and credence to AI and there are so many social issues tied to AI use that it is definitely something that we need to be concerned about.

But like social media and the evolution of media in general right now, it’s one of the things that we need to understand, accept, and learn to challenge and take ownership of in order to move forward together and create something better for humanity. I don’t think AI is a net good for humanity, but I also don’t think it’s entirely bad. I think we just need to have transparency in how it’s being used, and so, I’m doing my part in trying to be transparent.

What’s less straightforward is how to make AI development tools actually understand the site they’re building against. Not in the abstract “here’s a REST API endpoint” sense, but natively — the way a developer who’s worked on a codebase for years understands it.

The Model Context Protocol

The Model Context Protocol (MCP) is an open standard, originally developed by Anthropic, for connecting AI systems to external data sources and tools. Think of it as a universal adapter: instead of writing bespoke integrations for every tool an AI might need to use, you expose a standardized server that speaks a common language.

For Claude Code — Anthropic’s CLI for AI-assisted development — MCP servers register as native tools. When it works, you don’t ask Claude to “call this curl command.” You just ask it to look something up, and it does, using the same tool-calling mechanism it uses for reading files or running tests.

The goal here was to get Claude Code to natively understand jazzsequence.com — its post types, taxonomies, content structure, plugins, theme — so that when I’m building the Next.js frontend, Claude doesn’t have to guess at data shapes. It can just ask.

What WordPress Exposes

The foundation is the WordPress Abilities API — a custom API layer built on top of WordPress’s capabilities system. The Abilities API lets plugins register named “abilities” — discrete operations with defined input/output schemas, permission callbacks, and metadata. Think of it as a structured way to say “this site can do X” and describe exactly what X looks like.

The jazzsequence-mcp-abilities plugin registers 34 abilities covering discovery (post types, taxonomies, plugins, theme, menus, shortcodes, hooks, blocks, options, rewrite rules, custom fields, cron jobs), content CRUD (posts and terms), media (upload, update, delete), and operations (cache, cron, options). These are exposed through the MCP Adapter plugin — any ability with meta.mcp.public = true becomes a callable MCP tool. Plus meta-tools for runtime introspection, plus 29 Ninja Forms tools. 66 tools total.

What Had to Be Fixed

In theory, all of this was already built. In practice, nothing was registering.

I used Claude Code to write the connecting tissue (because part of the goal of the Next.js project is to experiment with Claude Code building a comprehensive site). I pointed it at all the available documentation for the Abilities API and MCP Server and even this video that Brad Williams did on the subject.

The first problem was in the plugin itself. The Abilities API requires abilities to be registered on the wp_abilities_api_init action hook — not plugins_loaded, which is where the plugin was initially hooking them. WordPress was silently rejecting every registration attempt because it happened too early in the load order.

The second problem was property naming. The Abilities API uses specific property names: execute_callback, permission_callback, input_schema, output_schema. Claude Code made assumptions about how these properties should be named and guessed wrong. Claude was using shortened versions — execute, permission, input, output — which WordPress would silently ignore. Not a single ability was registering. This had been broken across at least seven plugin releases. Fixing both got 34 abilities working.

The Proxy Problem

MCP servers can speak either stdio or HTTP. Claude Code natively spawns stdio servers. The WordPress MCP endpoint speaks HTTP with session management. The solution is a stdio-to-HTTP proxy: a Node.js script that Claude Code can spawn as a subprocess, translating stdio JSON-RPC into authenticated HTTP requests.

This proxy had its own chain of bugs. A race condition: Claude Code sends initialize and tools/list in rapid succession; the proxy was forwarding both concurrently, so tools/list fired before initialize returned with the session ID. Fix: sequential request queue. Next, there was a protocol version mismatch between what WordPress returns and what Claude Code sends. Fix: intercept and normalize. A timing timeout: two HTTP round-trips to WordPress took ~500ms, which exceeded Claude Code’s threshold for marking a server as having zero capabilities. Fix: respond to tools/list immediately from a local cache, establish the WordPress session in the background.

The hardest bug: Claude Code silently rejects a server’s entire tools list if any tool’s inputSchema properties contain enum or default keywords. Zero tools register, zero error messages, no indication of why. The WordPress tools use enum extensively. The diagnostic approach that finally worked: a minimal two-tool test server (working), then progressively swapping in real WordPress tools until it broke, isolating which schema keyword caused the failure. Fix: strip enum and default from property schemas in the proxy.

The Result

Claude Code now has 66 native tools for interacting with jazzsequence.com. At session startup it loads them from a local cache in under a millisecond; in the background it refreshes from WordPress and updates if anything changed.

From inside a Claude Code session working on the Next.js frontend, asking “what custom post types does this site have?” returns a live answer from the actual WordPress installation — not stale docs, not memory. All have Zod schemas in the Next.js codebase. The MCP integration means Claude can verify those schemas against real data without manually copy-pasting API responses.

Drafted by Claude (Sonnet 4.6) with final editing by a human.

I was officially named Interim President of The WP Community Collective this week.

When I joined The WPCC as a member last year, it was not too long after going to my first DrupalCon. DrupalCon Atlanta was enlightening for a lot of reasons. But most relevant to why I joined The WPCC was learning how the Drupal Association works and the relationship between the DA and Drupal core development.

Drupal and WordPress have very similar histories, and it’s really largely due to a software decision made 16 years ago that led to the disparity in widespread adoption between Drupal and WordPress. Drupal, despite occupying a fraction of the Open Source CMS usage as WordPress, is still growing and thriving and its community is still as passionate as it’s ever been. I came away from that event with the knowledge that there are lessons that WordPress can learn from how the DA operates, much of which I wrote about in that blog post.

It’s clear that the WordPress Foundation will never be an organization like the Drupal Association. And while I wish that could be different, that’s not the reality. So what is? Well, there just so happens to be this organization that is positioning itself to fill the gap to help build a contribution compensation model that is sustainable. And, oh hey, they also got a large donation from GoDaddy to help them get off the ground.

Honestly? When the possibility of expanding the Board of Directors came up, even when I first joined, I was hoping to be on it. I have a long history in the WordPress community and, since joining Pantheon, have been able to experience a little bit of the Drupal community as well. And I’ve been using and evangelizing Open Source for a long time, since before I found WordPress. My years in the community, my access to other, related communities, and my role at a SaaS provider that serves both seemed like they’d come in handy. And yeah, it looks good on a resume.

Conveniently, Sé Reed thought so, too, and invited me before I could ask. And, through a confluence of events, the Presidency landed on me. That part wasn’t planned.

As I wrote for the press release, I increasingly feel like Open Source is at risk of exploitation. It always has been, of course, but I believe that the rise of AI-assisted development inflates the risk exponentially. AI models will integrate Open Source packages and use Open Source code for reference freely, without attribution or citing sources. Developers who use LLMs in their development might be getting code from places they don’t even know, and have even less interaction with the original authors than before these tools existed.

And Free Open Source Software is no longer taboo with large organizations. We’re well past the days when Microsoft CEOs viewed Open Source Software as an existential threat — now Microsoft is using and supporting Open Source. This is great for Open Source! This is also terrible for contributors to Open Source. Because unless your project just happens to have a benefactor with deep pockets, it’s not unlikely that your code can and will get used by a much larger organization than yourself, without any direct benefit to you as a contributor or maintainer.

WordPress itself has struggled to overcome this issue. The solution as it has existed over the last many years has been “sponsored contributors” — that is, developers who are paid by their employers to contribute directly to WordPress core code. What do those employers get? It’s unclear, there’s no immediate direct benefit (beyond the warm, fuzzy feeling of ensuring the sustainability of the project — but it can be hard to justify that to a Chief Financial Officer who dismisses the value of financially sponsoring the funding of the Open Source Software their company uses; warm fuzzies don’t make balance sheets). It’s supposed that perhaps a company that sponsors core development can have some insight into what actually gets developed, but contributors aren’t often involved in feature or roadmapping decisions about what goes into WordPress core (which is somewhat different than how the Drupal Association works). The fact is, capitalism rules how Open Source Software is built. If it doesn’t fit inside a capitalist mindset, with a capitalist objective, it’s likely not getting funded.

And then there’s the question of “what counts” as contribution. I believe that you can contribute to the software ecosystem around WordPress (e.g. by building plugins or tools that are used for WordPress) without contributing code directly to core. But contributing code directly to core has always been seen as the gold standard for WordPress contribution, against which all other forms of contribution are assumed to be inferior. Contributing directly to core is not an easy path, nor is it necessarily an accessible path or an inclusive path for all potential contributors.

I also have direct experience with not counting. Did being mysteriously banned from WordPress have an influence on my decision to join a potentially controversial, counter-cultural non-profit organization poised to disrupt established norms in the WordPress ecosystem? Abso-friggen-lutely.

I love an underdog story. In many ways, fighting for compensation for Open Source creators is as anti-capitalist and anarchic as my chaotic high school self was. I fully believe that a better future can exist, and even exist inside and alongside the existing organizations and structures in the WordPress community. And through collaboration with those systems and a deep commitment to the human beings in this community, I am hopeful that The WPCC can be part of the solution. Because you know what’s really anti-capitalist and anarchic? Empathy and compassion.

Yesterday, the latest major version of WordPress was released. WordPress names each version after a jazz musician and this release was named after Gene Harris. Gene.

But naming a piece of software that I have worked in and around for 20+ years “Gene” does not make me think about new features. It only tangentially makes me think about WordPress at all.

Gene was my dad’s name. A name he got from his dad, my grandfather, Eugene. And this release is the first major release of a piece of software I’ve dedicated close to half of my life on since he died almost 2 months ago.

So, instead, I’m feeling hollow every time I read the release name. I could care less about what’s in this version of WordPress. I’m thinking about how I haven’t written anything about his death (besides a short post on Instagram). I’m wondering whether I just wasn’t paying enough attention to WordPress release names to notice that it alternated from last names (e.g. 6.7 “Rollins”, 6.6 “Dorsey”) and first names (6.5 “Regina”, 6.7 “Cecil”). I’m wishing that Ma.tt had chosen literally any other jazz musician for this release. I’m thinking about the time he asked me to build a website for him and, obviously, I did it in WordPress (this is where the tangential relationship to WordPress comes in, by the way).

I took a long hike after he died. I needed to feel something in my body. Process the pain and loss by feeling it in my muscles. And, in the process, I discovered for the first time parts of the place that he called home his whole life (and so did I, for the first half of mine). I “borrowed” a hat from him to take on the hike (so my head wouldn’t get sunburned). Now, despite not particularly being a SF Giants fan, I wear that hat to remember him. In that way that you feel like your loved ones are looking over your shoulder, I felt like we were experiencing the hike together. And we saw coyotes along the way.

We haven’t had a service yet for…reasons (that I’m not going to go into here), but I did write a eulogy when I was in the hospital with him and my mom. I’m not sure if I will ever get a chance to read it, so I’m posting it here. This is the Gene that I am honoring today. Not a jazz pianist I don’t know and never listened to.

My dad was always the “cool dad”. When I was old enough to care about such things, I always remembered how old he was by adding 21 years to my own age. Among my peers, he was always far younger – and by consequence, far hipper – than my friends’ parents. I never had the “ew, doing stuff with mom and dad” reaction as a teenager.

My childhood had a soundtrack. A mix of 70s rock, 80s modern rock and new wave. I experienced the world through my dad’s record collection. He once told me that before he got married, he wanted to be a DJ. Maybe I was fulfilling his dream a little bit, then, when, years later, I would start DJing.

Music is a language that we shared more than spoken words. A language I learned to speak before I knew the meaning of the words being sung (sometimes before I knew the actual song lyrics — as an adult I understand that indeed Iran is “so far away”, but A Flock of Seagulls was not singing about the Middle East; and while “only the lonely can play”, only bologna can not). My dad used to tell me about the concerts he took me to when I was still in my mom’s womb. And I’ve always wondered if those shows influenced how I still experience music.

As I learned this language of music from my dad, I began to want to share the new dialects I was learning with him. “Dad’s Father’s Day Tape” became an annual tradition and a way of sharing where I was in my life, even when the medium changed.

I realize now that a lot of things that make me who I am are extensions of who he was. You’re all here today because people gravitated towards him. He never sought out the spotlight, his spotlight just glowed naturally. But it was always one he was willing to share. He cared deeply about people, and in turn people cared about him. And he used humor to make people feel at ease and to not take things too seriously.

He and I never shared a lot of words about our feelings which is why I am thinking about this language of music. When I first flew out to see him after he went into the hospital, I was almost afraid I wouldn’t feel anything. That it would come later or that I would just feel numb. I didn’t expect that watching my dad tell jokes while he’s dying of cancer would be the thing that ripped my grief out of me and made me appreciate the person he was.

The hospice nurse was asking him questions like “do you remember what year it is?” and “do you remember what day it is?”, and he finally asked my dad “do you know where you are?” and after answering all the other questions, my dad said “Hell.” Then he added “that was a joke.” Regardless of what your faith tells you happens to us after we pass on from this world, there’s pretty much only one way to go from there.

This exchange made me appreciate how much he put others before himself and how much of that I pulled from him as well. He was still trying to entertain us even when it was hard for him to communicate.

I hate that so much time was stolen from him by this disease, that he’ll no longer be around to make us laugh. But I also know how to talk to him, how to listen to him and I know he’s not far away.

While sadly I didn’t do a lot of exploring, I had a number of takeaways from the event.

Organization

There have been a few posts reflecting on the event as I write this and I think it’s safe to say they all have (at least) one thing in common — the organizing team did a fantastic job. So much props to the Canadian WordPressers who put this together. It’s not easy to run a WordPress event — especially as a volunteer — and they deserve a ton of credit. This is an event that felt like it punched well above its weight class.

For my part, I was pulled in to act as an emcee. Besides emceeing at my own events (WordCamp SLC), this is the first time I’ve been formally named as such. It was a lot of fun, and I was happy to introduce and riff on the introductions for my friends Michelle Frechette and Dee Teal.

The vibe

I got a distinct feeling that WCEH had a higher concentration of what I’ve referred to as “movers and/or shakers” — people who are known either within the WordPress community specifically, or the broader internet software ecosystem. Bringing in speakers like Dave Winer (a fundamental contributor to the RSS protocol on which much of the internet and podcasting today operates) and Evan Prodromou (who helped create ActivityPub, the backbone of the open web Twitter/X alternative, Mastodon) made the event feel much larger in scope than just WordPress, even when both talks centered a lot on WordPress implementations and ideas. Other speakers, like Jill Binder, Carl Alexander and Brent Toderash, among many others, I could expect to see on the stage of a “flagship” event.

And that’s even before we get into Matt Mullenweg’s surprise appearance.

With (at least) three distinct voices who have had a hand in shaping the modern web in attendance, it felt a lot like PressConf did (it didn’t hurt that opposite my emceeing spot was PressConf’s founder and lead organizer, Raquel).

My WCEH Wapuu video (and possibly my recent pro-Wapuu post) and even the Sora AI experiment (where I asked Sora to make me rap about WordCamp Canada, far more effectively than I could actually rap about…anything) seemed to do the job they were intended to do. While I don’t want to make myself out to be focused on social media success or being a “content creator” (even though that’s exactly what I am ), I did come up with the idea for and started creating those shorts (and the Community + Code podcast) as a way to “elevate my personal brand”. I say that it seems like it’s been effective because even if people didn’t recognize me, they recognized the videos (and had good things to say about them). And Wapuu met some new friends at the event.

But the biggest “vibe” takeaway I had was the feeling of community. A lot of the folks in the WordPress community I’ve known for years. Some I’ve only known of. And still others I’ve only met (virtually) recently. But, being at WCEH felt like I was surrounded by friends. And this was far more so than at WordCamp US a couple months ago. Maybe it’s a Canadian thing, or maybe because the size of the event was 1/6th the size of WCUS in August, but I left feeling very deeply appreciative of the connections I’ve made in the time I’ve been in this community and felt way more enthusiastic and positive about the WordPress project and ecosystem as a result.

The event

Part of my job as a Developer Advocate at Pantheon is to be the face (or one of) of Pantheon for the community. That means that, far more so than in my previous engineering roles (at Pantheon and elsewhere), when I go to events like this, I’m working. Don’t get me wrong, “working” in this context very much means having conversations with people and making connections. And I enjoy doing that. But that means that often, that facetime takes precedence over other things.

As such, I didn’t make it to everyone’s talk, but I was pretty happy with those that I did attend. It’s a bit of a “known issue” that at a certain point in your WordPress journey, you get to a point where you aren’t learning much at events like this anymore. As such, those sessions I do go to tend towards the “soft skills” side of things rather than technical talks, or I go to support my friends or former colleagues. But I definitely came away from this camp feeling like I learned stuff. And beyond the talks I went to, there were a number that I didn’t make it to that I would have had I not been caught up in the “hallway track”.

And the hallway track was really good. Again, there were a lot of interesting people in attendance which made it easy to have interesting conversations with people. And easy to find some folks to recruit as guests for my podcast.

While I didn’t get to really explore much of the Carleton University campus or Ottawa generally, the view outside was also very pretty.

Finally, while I won’t pretend to agree with the #WPDrama between WordPress’ founder and a private-equity-owned competitor to the company that I work for, I was interested that Matt decided to do a live Q&A at such a small event. And while I don’t feel like he adequately answered the question I asked at the town hall, from the perspective of earning the ticket that took me to a place where you can, apparently, get poutine from Pizza Hut, I was able to:

1. name my company
2. talk about drupal
3. appear (sort of) on a video asking Matt a question

which I take to be a success.

(I did not have the poutine, which was not a success…not that I expect Pizza Hut to have vegan-friendly poutine…)