In 2008, I released Blogroll Links, a WordPress plugin for displaying curated lists of links inside posts and pages. I added its shortcode interface in February 2010: a marker such as [blogroll-links categoryslug="friends"] tells WordPress where to insert the list. My original post records that change. The small plugin served a larger ambition: people owning portable identities and relationships on the web. In November 2025, I returned to it with Claude Code to modernize software I still wanted people to use.
I built it because I believed people should control their own online presence. XFN, the XHTML Friends Network, gave that idea a concrete form. An ordinary HTML link could carry relationship information. With rel="friend met", you declared that the person whose site you linked to was a friend you had met in person. With rel="me", you declared that another URL represented you.
Those declarations made relationships readable by software without requiring every site to keep its own private copy of the same social graph. Google launched its Social Graph API in February 2008 to discover public connections, including XFN links. For a developer who wanted websites to connect on their owners' terms, that was a promising direction.
Facebook and Twitter became the places where much of social networking happened. Google announced the API's retirement for April 2012, citing limited adoption. The idea of people owning their identities and connections stayed with me.
The open web gave the maintenance work a purpose
By the time I returned to the plugin in November 2025, the Fediverse had given that aspiration another practical expression. Services such as Mastodon and Pixelfed let people participate through independently operated servers.
XFN puts relationship metadata on hyperlinks. ActivityPub defines protocols for clients to interact with servers and for servers to exchange social content. The Fediverse realizes the kind of open, portable social web I wanted XFN to help make possible. The Social Web Foundation launched in September 2024 to support that work.
WordPress sites can participate through the ActivityPub plugin, with the plugin and site configured for federation. That connection matters to me: people can use their own publishing sites as a place others can follow.
The IndieWeb practice called POSSE, Publish on your Own Site, Syndicate Elsewhere, offers another route. Publish at an address you control, then distribute copies or share links elsewhere with a path back to the original. A blogroll on your own domain provides the owned-publication part. It does not, by itself, do the syndication. Own your content. Own your relationships. Participate in networks without surrendering control.
Blogroll Links still had a useful job: publish links and relationship declarations from a site's own database. Its implementation needed security, PHP compatibility and admin-form work before I was comfortable recommending the new release.
Why I still build
Throughout my career, as CTO of The New York Times, Chief Product and Technology Officer at The Wall Street Journal and Hearst Magazines, and earlier leading technology for Condé Nast’s digital brands including Reddit, I’ve led teams of hundreds of engineers building systems serving millions of users. I’ve never believed that seniority means distance from the work. The best product and technology leaders I know maintain what I call “technical currency”: the ability to understand systems from the inside.
Building and maintaining software changes the questions I ask about architecture. I can examine where a design accepts input, where it stores state and how it fails. When an engineering team explains a difficult tradeoff, I can reason about the implementation with them. That ability needs practice; reading status reports doesn't provide the same contact with the work. Keeping that contact helps me ask better questions and recognize problems earlier.
A neglected plugin was a useful place to apply synthesis coding, the hands-on craft of directing and verifying AI-assisted software development. Claude Code assisted with implementation while I set the requirements and reviewed the result. The broader organizational discipline is synthesis engineering.
The four principles I used to organize this 2025 case were:
- Human architectural authority. I set the security requirements, coding standards and compatibility targets, and made the decisions about what entered the release.
- Systematic quality standards. AI-generated code received review and testing. Its origin didn't reduce the standard it needed to meet.
- Active system understanding. The people responsible for the plugin needed to understand its code well enough to maintain and debug it.
- Iterative context building. Written constraints and architectural notes made decisions available for subsequent work. Each maintained explanation gave the next task a better starting point. Consulting that accumulated context and checking the implementation against it made later work more effective.
The framework expanded to five principles in April 2026. The current framework describes that evolution, including transferable knowledge and the broader self-improving-system principle. The list here preserves the four-part framing of the original case.
Exploratory coding with AI has its place. For this plugin, the work had a defined release target and users who would rely on the output. Synthesis coding gave me a practical way to build secure, maintainable software at the pace of AI-assisted implementation while keeping architectural decisions and verification in human hands.
The changes in Blogroll Links
The older code reflected a different era of WordPress development. It assembled database queries with sprintf(), assumed a wp_ table prefix, and accessed the first category result without checking whether one existed. Its generated HTML inserted bookmark fields directly into markup. The admin form displayed settings it didn't reliably save.
Standards evolve, and I’ve evolved with them.
Working with Claude Code, I set WordPress 6.x and PHP 8.x as modernization targets, with PHP 7.4 as the declared minimum. The plugin already used GPL version 2 or later; that license remained in the release. WordPress coding conventions, clearer PHPDoc documentation and tests were part of the work.
The query change is visible in the versioned source. The earlier lookup interpolated a category slug into a SQL string:
$sql = sprintf("SELECT term_id FROM wp_terms WHERE slug = '%s' LIMIT 1", $category_slug);
The replacement sanitizes the slug and passes it as a value to WordPress's query-preparation API:
$sql = $wpdb->prepare(
"SELECT term_id FROM {$wpdb->terms} WHERE slug = %s LIMIT 1",
$category_slug
);
The value is supplied separately from the SQL format, so the API handles its quoting and escaping. The table comes from $wpdb->terms, so the lookup no longer assumes every installation uses the wp_ prefix. The helper also returns an explicit failure value when it cannot find a category, allowing the caller to render an error message.
The admin-form work addressed a different boundary. The new handler checks current_user_can( 'manage_options' ) before exposing its settings. On submission, it separately verifies a nonce tied to the settings action. A WordPress nonce helps protect against cross-site request forgery, where an attacker tries to make a user's browser submit an unwanted request. It is not proof that the user has permission. WordPress's nonce documentation explicitly separates nonce checks from authorization.
The form then sanitizes incoming values, validates the allowed sort field and direction, saves each setting and loads those values back into the controls. The old implementation used a checkbox helper in text-field values and updated an unrelated new-window option; the new form's field names and stored options correspond to the settings it presents.
For rendered links, Claude Code added context-appropriate escaping: esc_html() for visible text, esc_attr() for relationship and target attributes, and esc_url() for URLs. These functions have different output contexts. Applying one indiscriminately to every value would miss the reason for the change.
PHP compatibility work included initializing the generated output, supplying option defaults and handling empty lookup results. The changes also added PHPDoc descriptions of parameters and return values, so a future maintainer could follow the same paths.
What I checked
I set up a local WordPress environment with wp-env, created link categories and sample links, checked the shortcode output, and confirmed that the admin panel saved and loaded its values. I also checked for PHP warnings with debugging enabled.
The checked-in wp-env configuration selects WordPress 6.7 and PHP 8.2. The release metadata declares WordPress 5.0 and PHP 7.4 as minimums and lists WordPress 6.7 as tested up to.
Claude Code also created a PHPUnit suite. Its checks cover hook registration, invalid and empty category slugs, output structure and other plugin behavior.
The result of that weekend was version 3.0.0, with inspectable changes at the query, form and output boundaries. I completed in a few hours work that would otherwise have occupied days. I wanted to give this useful tool another decade of service.
For a maintainer applying the same approach, these boundaries suggest concrete checks. Try a missing category as well as one with links. Save a changed setting and reload the form. Exercise a rejected request, not just a successful submission. Check rendered names, descriptions, attributes and URLs with inputs appropriate to each context. Run the plugin in the environment in which you intend to use it.
September 2026 compatibility note: The WordPress directory warns that the plugin has not been tested with the latest three major WordPress releases. Test it with the WordPress and PHP versions you intend to use.
Owning a list of relationships
A blogroll on your own domain keeps the published list under your control. If a social platform changes its API, that does not force you to rewrite the software that serves your list. If Twitter (or X, or whatever it's called next week) changes course again, you still choose which links appear on your site.
XFN adds useful meaning to those links. rel="friend" and rel="colleague" express different relationships. rel="me" declares that another URL represents the same person. A declaration can be used by software that checks relationships, but the attribute alone doesn't authenticate ownership of an account.
I still believe in the idea that drew me to this work in 2008. The Fediverse shows that the model works, and ActivityPub shows that open protocols can support a working social web. WordPress’s ActivityPub integration brings the tools people already use into that ecosystem. Blogroll Links contributes its own piece: a person can curate links, publish relationship information and keep that record on a domain they control.
Tools like this are unglamorous. They don’t have venture funding or growth metrics. They help people own their piece of the web.
The WordPress plugin directory and source repository keep the project available for inspection. My WordPress contributor profile is part of a longer history of working on open source alongside product and technology leadership.
Bringing that work back to leadership
Economic pressure and AI-assisted implementation are compressing the layers between senior leaders and technical work. Leaders who maintain technical currency can evaluate architectural proposals from experience and recognize problems they have encountered in their own work. Combining strategic leadership with maintained technical capability is what I think the best product and technology executives will embody going forward.
During the modernization, I made architectural decisions, reviewed security-sensitive code, checked functionality and published a release through WordPress.org. It took a few hours. Then I returned to the strategic work that occupies most of my time. Synthesis coding makes that direct engagement practical at a pace that leaves room for the wider responsibilities of leadership.
For a leader evaluating AI tools, that experience supplies better questions than an abstract discussion about how much code a model can produce. Which constraints did the tool receive? What did the implementation change? Where did the reviewer check the result? Which parts of the claimed outcome still depend on an untested environment?
If you want the wider method, the professional-practice article, framework and technical workflows develop it beyond this case.
If you’re a WordPress user who cares about the open web, use Blogroll Links to curate links on your site and add XFN relationships. Own your corner of the web.
For your own work, consider a tool from your past that still serves a purpose. Identify the behavior worth keeping, the assumptions that have aged, and the evidence a new release would need. Returning to something you built years ago can put your current judgment to work on a problem you know deeply.
The open web is worth building. Synthesis coding makes more building possible. Seventeen years later, I still want to do both.