The 5 Levels of Agentic WordPress

This was where much of AI-driven programming or building began. Often we don’t even realise it’s there anymore (especially in places like Gmail).

AI-powered autocomplete first shipped in an IDE back in 2018, and was further made popular by GitHub Copilot in 2021. By reading what you’ve coded so far, it’s quite easy for a system to predict what comes next (and that’s the crux of it). It appears as ghost text, and once you press Tab, it transforms into completed words. This pattern has become popular in many of our daily tools.

Whilst it’s a nifty little feature, it doesn’t have the context of an entire project or codebase; it’s simply operating within a fragment or a page. It’s also become expected at this point, or dare I say, somewhat insignificant in light of the agentic developments we’ve witnessed since.

Nowadays I wouldn’t even necessarily consider this the true “level one”, but I think it’s become popularised due to its origin story.

Let’s move on from an editor suggesting what comes next in a function, to one that writes the entire function.

When we first got access to ChatGPT and it started becoming insanely popular, we could suddenly ask it to create entire functions or plugins.

It was comical at first, as we’d have to split it up into multiple threads or multiple messages to assemble larger fragments of code together, because the chat interface at the time was unable to output that much in one go. We’re completely past that point, coding agents will absolutely write entire apps for you in one go.

If you’re still copy-pasting snippets, this is your highest-leverage play today (and it’ll only take a few hours to figure out):

Frontier models and chat interfaces also became stronger. They allowed for longer messages, allowed for uploading files and context, or they’d search documentation themselves and provide you with continuously better code to copy-paste.

The real game changer came with the introduction of coding harnesses such as Claude Code or Codex: terminal-style interfaces sitting on top of your code in the file system. All of a sudden, context was provided. Code didn’t have to be handed to you and pasted somewhere else; it could just edit the files directly, and from there we started moving extremely quickly.

This is where most people stagnate: they’re amazed by what the tools can now do, but don’t necessarily push further to see how much more they’re even capable of.

I think, at a minimum, graduating from this level requires knowing how SKILL.md files work and how they can benefit the way you like to work, as well as being able to give your coding harness access to the various MCPs you may use.

The tooling for this has quietly fallen into place: an agent on any of your devices can now talk to a WordPress website directly, with plain English as the programming language.

The Abilities API is almost an agentic version of the REST API, allowing agents to easily communicate with your website. Not only to read and understand parts of your website, but to effectuate changes too.

The MCP adapter is the highway the Abilities API runs on.

As the Abilities API is still new, not everything a plugin or WordPress offers is available to agents today. The workaround is WP-CLI, assuming you have the ability to SSH in. SSH is something agents understand very well and can create recipes (aka SKILLs) for.

Now it’s getting fun, putting agentic on autopilot.

The idea of agents going out there and doing work by themselves sounds daunting to set up, but if you’ve been following along, it’s basically just the components you’ve learned so far, with the only difference being they’re on a schedule. And that schedule can be as simple as a cron job.

For example, you may have a SKILL.md that does some kind of data hygiene on your website, connecting to it through the Abilities API. That skill is wrapped in a prompt that sits in one of the many schedulers that exist for agents out there.

Getting this far is already a superpower, as you’re able to delegate a lot of repeatable and somewhat predictable work to agents. This is work that was always a bit too wishy-washy for Zapier or IFTTT, but the outcomes aren’t so creative or novel that an agent can’t handle them well. That sweet-spot is perfect for agents as it produces consistent outcomes despite differing inputs.

By mastering level four, your website essentially comes to life.

The previous level, level four, is essentially the Truman Show: everything is alive, but it’s a sealed world. It doesn’t have the benefit of external data, be it behavioural signals, technical signals or otherwise, so nothing from the outside ever changes the decisions it makes.

Level five is that same setup dropped into the real world, where outcomes push back and inform your website how to react accordingly. Below is an example of that: our agent running multiple rounds of A/B testing to figure out how to increase the conversion rate itself, as quickly as possible.

And this is where your agentic setup starts mimicking nature: evolving itself to perform better. This is the only level that deviates from the built-in commands of the coding harnesses and other popular frameworks or packages, as you’ll have to create self-learning mechanisms for optimisation. Well, not you directly, obviously; frontier models are all extremely capable of handling such tasks.

The levels are not hard, or said differently, they’re a lot easier than other things we’ve had to learn in the past (related to websites and programming). A lot of it simply works on natural language. You could probably select all → copy all the text on this page and feed it to your Claude Code or Codex, so that it designs you a short curriculum (or focus on any part you want).

We want to think it’s hard, because these things used to be hard. However, the playing field has changed, and here you’re rewarded with functional apps and tools very quickly. Code is not the blocker.

The complexity comes later when you’re trying to manage tokens, context, reduce errors, increase output quality. Or you’re working on genuinely large or complex projects that require more thought, creativity and perseverance.

Assess where you are, and level up today.