WordPress before and after the AI revolution

Before the current AI wave, WordPress craft looked like memorizing APIs, reading core tickets, and typing faster than your doubts. After it, craft looks like knowing which generated answer is confidently wrong.

That shift matters because WordPress teams are under pressure to ship more with the same headcount. AI can help. AI can also industrialize mediocre architecture if nobody is accountable for taste, security, and domain judgment.

I lead developers. I also still write code. So I care about both sides: speed and what still has to stay human.

What changed for working WordPress developers

The old bottleneck was often blank page friction. How do I scaffold this CPT registration? What is the cleanest way to enqueue this script only on one template? Why is this WP_Query ignoring my meta query?

AI collapses that friction. Boilerplate appears in seconds. Draft docs appear in seconds. A first pass at a migration script appears in seconds.

The new bottleneck is verification. Does this respect capability checks? Will this break multilingual permalinks? Did the model invent a filter hook that never existed? Is this “optimization” deleting a script your CRM form still needs?

In other words, AI moved the hard part from generation to judgment.

Where AI is a scalpel on real projects

Used narrowly, AI is excellent for:

Drafting plugin stubs and test outlines
Explaining unfamiliar code during handovers
Turning a messy ticket into a clearer acceptance checklist
Suggesting regex or WP-CLI commands you then verify
First pass copy for admin UI labels and internal docs
Generating alternate image concepts when editorial needs volume

On Study International, we built an AI image generator plugin with multiple models and fallbacks. That was not “AI because AI.” Editorial needed image variations at scale. The plugin earned its place because it sat inside a real workflow, with failure handling when a model timed out or returned junk.

Where AI becomes malpractice

Handing AI the patient looks like this:

Merging generated code nobody reviewed
Pasting production secrets into a prompt
Letting a model redesign your data model from a paragraph of vibes
Using AI output as the security review
Shipping performance “fixes” that disable scripts without measuring conversions

WordPress sites are full of soft edges: hooks, race conditions, caching layers, third party scripts, editor habits. Models are good at the median example. Production is rarely the median example.

Team lead rules that keep AI useful

When I lead a team, AI is allowed. Unreviewed AI is not. A few rules that have saved us pain:

No secrets in prompts. Ever. Use redacted examples.
No merge without a human review that could explain the change. If the author cannot defend it, it does not ship.
AI can draft tests. Humans decide coverage. Generated tests that assert nothing are theater.
Measure outcomes. Especially for performance and SEO claims.
Juniors get acceleration. Seniors keep accountability. Mentorship means teaching verification, not just prompting.

These rules are not anti innovation. They are how you keep velocity from becoming a liability.

Before AI: the skills that still matter

The valuable WordPress engineer was never the person who remembered every function signature. It was the person who could answer:

What business problem is this ticket dancing around?
Which layer should own this logic: theme, plugin, or integration?
What happens on mobile when the third party tag manager balloons?
How do we deploy this without a Friday incident?

Those skills did not expire. They got more important, because generation got cheap and architecture mistakes got easier to mass produce.

After AI: a better default workflow

My current default looks like this:

Write the problem and constraints in plain language first.
Use AI to draft options or boilerplate.
Throw away anything that smells like generic blog code.
Implement the smallest clean version in the real codebase.
Review for security, performance, and editor experience.
Document the decision so the next person is not guessing.

Notice what stays human: problem framing, taste, review, and documentation. AI is invited into the middle, not crowned at the ends.

What clients should ask vendors now

If you are hiring a WordPress team in an AI saturated market, ask better questions than “do you use AI?”

Where do you forbid AI in your process?
Who reviews generated code?
How do you handle secrets and client data?
Can you show a case where AI sped up delivery without lowering quality?
What do you still insist on doing by hand?

The teams worth hiring will have answers that sound operational, not magical.

The point

WordPress before AI rewarded memory and grind. WordPress after AI rewards judgment under speed. I use the tools. I like the tools. I will not pretend a model understands your university CRM integration, your Woo sync edge cases, or your editor who will paste a whole PDF into a content field at 4pm.

AI is a scalpel. Handing it the patient is still malpractice.