Adios WordPress

For almost twenty years, Memory Leak lived on WordPress.com.

That sentence still feels strange to type — because the blog is older than WordPress. In college I built the first version by hand: raw HTML, uploaded to a host, no CMS, no admin panel, no safety net. I wrote about Paul Reed Smith guitars and the Dave Matthews Band and whatever else seemed worth putting on the internet. It was clumsy and mine.

WordPress entered the picture in March 2007, when I published a post I still find amusing — I Can’t Believe I’m Blogging — and predicted I’d get bored in three to six months and move on to a podcast. I was wrong about the timeline. WordPress was the obvious choice back then: cheap, durable, familiar, and good enough for a personal site that might occasionally go viral when I wrote about the Giants winning the Super Bowl or ranted about the iPhone. It carried hundreds of posts, survived platform redesigns, plugin churn, and more than one “I’ll get back to blogging” drought.

WordPress did its job. I am not here to bury a platform that still powers a huge slice of the web. According to W3Techs, WordPress runs roughly 41% of all websites and about 59% of sites that use a known content management system. That is not a niche tool — it is the default gravity well of the CMS web. Millions of publishers run perfectly well on WordPress today, and many always will. For a lot of them, the login-and-publish workflow is exactly the right trade.

The irony is not lost on me. This post is not a verdict on their choice. It is an account of mine.

But the web I am building for now is not the web WordPress was designed to serve.

SEO, AEO, and GEO — and why the foundation still matters

For twenty years, SEO — Search Engine Optimization — meant making your site crawlable, fast, structured, and trustworthy enough that Google (and friends) would rank it and send humans to click. That work is not obsolete. It is the foundation. Fast static HTML, semantic markup, clear headings, canonical URLs, honest metadata — still the price of admission.

What changed is what happens after the crawl.

AEO — Answer Engine Optimization — is the discipline of being understood, summarized, and cited correctly when someone asks ChatGPT, Perplexity, Google AI Overviews, or the next answer engine a question your content could answer. Not gaming the model. Being legible: who wrote this, what is the thesis, what is the evidence, what is the TL;DR.

GEO — Generative Engine Optimization — is the broader frame: optimizing for generative surfaces where AI synthesizes answers from many sources — chat, copilots, embedded assistants, agent browsers. The names blur. The requirement does not: your site must survive being read, compressed, and repeated without losing meaning or attribution.

What this means for every website — not just mine:

• SEO gets you into the corpus. If agents and answer engines cannot fetch clean HTML, you are not in the game.
• AEO/GEO determine how you are represented once you are in — quoted accurately or garbled, credited or anonymous, chosen or ignored.
• Structure beats tricks. Original information, consistent authorship, schema that corroborates the page — not a parallel “AI version” of your site.

Google and every serious strategist I have studied converge here: do HTML well first. The agentic layer is not a separate track. It is what good publishing becomes when machines read alongside humans.

A hard truth for marketing teams

This shift is going to expose a lot of marketing organizations that have been just getting by.

For years, many teams could survive on a familiar playbook: WordPress or a headless CMS, an agency retainer, a keyword spreadsheet, a content calendar, monthly traffic reports, and a plugin that claims to “handle SEO.” That was often enough when the job was to rank in ten blue links and hope humans clicked.

AEO and GEO are not a new coat of paint on that playbook. They sit on top of technical foundations most marketing teams were never asked to own — and often were actively shielded from: semantic HTML, structured data, canonical entity graphs, crawl and render behavior, redirect integrity, performance budgets, how RSS and sitemaps actually ship. The vocabulary alone — schema @graph, Person @id, prerendered JSON-LD — is a foreign language in most marketing standups.

That is not an insult. It is a capability gap. And agents make the gap visible fast. When the evaluate → pick → do arc runs in a context window you never see, there is no monthly dashboard smoothing over weak structure. Either your site is legible, attributable, and trustworthy — or a competitor’s is, and the agent never mentions you.

Teams that thrive will stop treating the web as a campaign surface and start treating it as infrastructure: marketing and engineering at the same table, shared definitions of authorship and facts, content that is original enough to cite, and architecture that does not require a developer to reverse-engineer the theme every time someone asks “can agents read this?”

Teams that keep coasting will wonder why traffic still looks fine while mindshare quietly migrates to brands that built for the machine reader. The metrics lag. The shift does not.

I say this as someone who has spent years in the room with both sides. The future belongs to marketing leaders curious enough to learn the technical layer — or honest enough to partner deeply with people who already have. There is not much middle ground left.

The agent arc: evaluate, pick, do

If you’ve read my Reflections on AI series — the arc from software eating the world through accelerating returns, context and memory, and the stochastic era — you know the through-line: AI is not a feature bolted onto software — it is a force reshaping how software is built, consumed, and understood. Leaving WordPress is part of that same story. On the web, the force shows up in three stages — and we are already moving through them.

1. Agents evaluate. Research assistants, comparison bots, and answer engines read your site (and your competitors) to summarize options, extract claims, and form judgments. This is happening today when someone asks “what does Renato write about AI?” or “compare these two approaches to migration.”

2. Agents pick. The next step is recommendation with intent: which product, which vendor, which author, which path — chosen on a user’s behalf from what the agent understood. Your site is not just visited. It is shortlisted or discarded in a context window you never see.

3. Agents do. The end state is action: book the flight, file the form, subscribe to the newsletter, deploy the config, merge the PR. The web stops being a brochure humans click through and becomes an API surface agents operate. First-class citizenship is not a metaphor. It is the difference between being referenced and being usable.

Agents will be first-class citizens on every digital property I build. Not as marketing. As architecture. Human readers and agent readers should encounter the same facts, the same structure, the same authorship — not a theme shell optimized for eyeballs with the real content buried underneath.

WordPress was built for a human publishing workflow: log in, write in an editor, hit publish, hope the theme cooperates. That model made sense for a long time. It does not match the evaluate → pick → do arc — not for how I want to build this site.

This is not a blog upgrade

The mistake would have been to “modernize WordPress” — swap themes, add a headless layer, keep paying for hosting, and call it done. That would have been a coat of paint on a house whose foundation doesn’t match the lot I want to build on.

Instead, I treated renato.mascardo.com as what it actually is in my head: a developer sandbox with a blog at /blog/*, not a blog that happens to have a domain name. Full code control. Git-driven workflow. Room for experiments — optional API routes, structured entity data, maybe an MCP endpoint later — without dragging a PHP runtime and MySQL instance along for the ride.

The stack is deliberate and boring in the best way:

• Astro 6 — static HTML at build time, zero client JavaScript by default
• Vercel — preview deployments on every pull request, production on merge
• Content in Git — Markdown with typed frontmatter, not a database
• No WordPress after cutover — export, migrate, redirect, cancel hosting

We migrated 235 posts, rewired 253 redirect routes so old permalinks still land in the right place, and rebuilt the visual layer on a QuietPages-inspired design — dark, restrained, readable. The celestial hero and gold accent are polish. The architecture underneath is the point.

Agents as first-class citizens

The design thesis I locked early: agent-friendly means well-designed HTML.

Every public page ships with semantic landmarks: one <h1>, logical heading order, an <article> that means what it says. Blog posts expose a TL;DR in the page and in JSON-LD as abstract. Authorship points to a canonical Person entity declared once in person.ts and referenced by @id everywhere else — no drift between About, footer, schema, and bylines. RSS serves rendered HTML, not raw markdown ghosts. Images require alt text. The sitemap and robots.txt know what’s public.

That is what agentic readiness means in practice: structure you can trust, identity you can resolve, content that survives copy-paste into a context window without losing meaning.

Legacy CMS platforms can approximate some of this with plugins and patience. They were not born there. I did not want to retrofit agent literacy onto almost twenty years of theme debt. I wanted it in the templates from Phase 1 — before the migration finished, not after.

What WordPress gave me — and what I am taking with me

WordPress lowered the floor for publishing when the floor needed lowering. It kept Memory Leak alive through seasons when building was more interesting than writing. It hosted family updates, hot takes, book notes, stroke recovery journals, ski trip reports, and a long run of AI reflections that only make sense in hindsight. I am grateful.

I am also honest about the friction that finally pushed me over the edge: export anxiety when you realize your archive lives in someone else’s garden; theme and plugin layers you do not fully control; the slow drift between what you write and what actually ships to the wire; hosting that keeps ticking because the site is hosted, not owned. None of that is catastrophic. All of it is cumulative. And the more I thought about building on this domain for the next twenty years — not just blogging, but building — the less sense the dependency made.

What I am taking with me is the work: the archive, cleaned and version-controlled, every old URL redirected to its new home under /blog/. What I am leaving behind is the platform.

Done with WordPress is a decision, not a drama.

Built with an agentic software development lifecycle

Here is the part that still surprises me: I would not have had the time to build this site entirely on my own.

Not at this quality. Not on this timeline. Not while running everything else in life and work that competes for the same hours.

This rebuild — migration scripts, redirect verification, security audits, design port, content QA, the new authoring workflow you are reading this post through — ran on a fully agentic software development lifecycle. Human intent and judgment at the center; AI agents for research, implementation, review, and iteration across the stack. Plan in Markdown. Code in Git. Preview on every branch. Verify before merge.

That is not cheating. It is the new normal for builders who pay attention. We are in a new era of building — the same way the web made everyone a publisher, agents are making small teams feel like large ones. I lived it on this project. It was exhausting and amazing in equal measure.

The site you are reading is evidence, not theory.

A new home for building

The DNS cutover is behind us. WordPress is read-only history. Hosting gets cancelled. Memory Leak is still Memory Leak — same voice, same themes, same appetite for long arcs and occasional hot takes — but the site underneath is mine in a way it never was before.

This is not just a new blog theme. It is a new home for building: a place where I can write, ship experiments, expose structured data, and extend the surface area over time without asking a CMS for permission. If you have followed this site since 2007 — or since the earlier hand-coded days — thank you for reading through every pivot, drought, and reinvention. If you are arriving now, welcome. You are seeing the version built for SEO, AEO, GEO, and the agentic web that follows.

The writing continues. The plumbing changed on purpose.

Adios, WordPress. You carried the blog a long way.

Memory Leak was never just a URL. It was always a habit — think, write, publish, repeat. That habit has a better address now.