A practitioner’s honest report from the last six months.
There is a lot of noise right now about AI and what it means for web development. Most of it falls into one of two camps: breathless enthusiasm from people who discovered ChatGPT last month, or existential panic from people who have confused their own uncertainty with an industry verdict.
This is neither. This is what actually changed in my stack, my workflow, and my reasoning over the past six months — and what did not change at all.
The evolution nobody is talking about clearly enough
If you have been around WordPress long enough, you remember what building a site used to look like. You had elements — pre-made building blocks in Oxygen, Bricks, Elementor — and you clicked, inserted, and manipulated them into something that resembled what the client wanted. Flexible, yes. But still fundamentally constrained by the blocks themselves. Getting to the actual HTML and CSS underneath meant either fighting the builder or writing custom code outside of it.
Etch changed the direction of that conversation. The premise was direct: write HTML and CSS the way they were designed to be written, without the abstraction layer getting in the way. Not back to 2003, but back to the actual craft — closer to the metal, less divception, less bloat.
AI takes that direction further. And this is the part most people are either missing or overstating.
AI does not replace the understanding of HTML and CSS. It amplifies it. If you know what good markup looks like, what a clean component structure should be, what accessibility requires — AI becomes a very fast pair of hands executing on your reasoning. If you do not know those things, AI gives you plausible-looking output that you cannot evaluate. That is not a workflow upgrade. That is technical debt at machine speed.
The practitioners who are genuinely benefiting from AI right now are the ones who built the craft knowledge first and are now using AI to move faster within it. Not the ones hoping AI will substitute for knowledge they never acquired.
Not every site needs WordPress anymore
This is the shift I did not fully anticipate, and it has had real consequences for how I approach new projects.
WordPress is an extraordinary tool. It is also a significant attack surface. Plugin vulnerabilities, authentication endpoints, database exposure, xmlrpc, the admin panel — every WordPress site carries a threat model that requires active management. For sites that genuinely need what WordPress provides — dynamic content, user management, e-commerce, complex CPTs, editorial workflows — that trade-off is worth making. You manage the security stack properly and you get a powerful, flexible platform.
But a lot of sites do not need any of that. They need reliable, fast, maintainable delivery of content that does not change hourly. And for those, a static site — Astro, in my case — eliminates entire categories of attack surface while being faster to deploy, cheaper to host, and simpler to reason about.
The question used to be: “should this be WordPress?” Because WordPress was the default and deviation required justification. The question now is more honest: “what does this site actually need to do, and what is the right architecture for that?” AI-assisted reasoning has made that question easier to answer precisely and document clearly — which means the client conversation changes too. You are no longer asking them to trust your instinct. You are walking them through the reasoning together.
That is a different kind of professional relationship. Better, in almost every respect.
What “using AI in your workflow” actually means
The surface-level version of AI workflow integration looks like this: you get stuck on something, you open a chat window, you ask a question, you get an answer, you close the window, you continue. Useful. Better than a Google search in many cases. But fundamentally episodic — the AI has no context, no continuity, no understanding of your business, your clients, or your standards.
That is not what I am doing anymore.
What replaced it is a cohesive personal AI infrastructure — a system where the AI has persistent context about how I work, what my clients need, what my standards are, and what problems I am currently solving. New ideas, client information, content, technical challenges — everything gets fed into the same system, and the output reflects that accumulated context.
The difference is between asking a stranger for directions and working with someone who has been on every project with you for the past year. The stranger might give you correct directions. But they do not know that you have a deadline, that the client is difficult about scope, that this particular problem has come up twice before, or that the last time you took that route it cost you three days of rework.
Six months ago, that kind of AI integration was not practically achievable for a solo operation. Now it is. That is the actual shift — not AI as a party trick, but AI as a genuine extension of how you think and work.
On the fear
I produced a podcast episode recently called Stop Writing Obituaries for Tools You Never Understood. The thesis: the “Tool X is dead” genre of content is the laziest form of professional commentary, and the most damaging — because it conflates one person’s narrow experience with a universal verdict, and it reaches people at the exact moment they are most susceptible to acting on bad information.
The “we are cooked” narrative is the same pattern, scaled up. It is an industry-level obituary written by people who are either genuinely frightened or have found that fear-adjacent content performs well. Sometimes both.
I am not frightened. I am busier than I was twelve months ago, working on more technically interesting problems, with a clearer sense of what I am good at and what I will not take on. The workload has not collapsed. The craft has not been automated away. The clients who understand what they need have not stopped needing it.
What has changed is that the bar for vague, undifferentiated execution is lower. If what you offer is “I build websites” with no further precision, that is a harder position to hold than it was two years ago. The tools are more accessible, the execution is more commoditised, and clients are more aware of what is achievable without a professional.
But that was always going to happen. AI accelerated it, but did not invent it.
What the actual advantage looks like
The practitioners who will do well in the next few years are not the ones with the best prompts or the most AI subscriptions. They are the ones who:
Know their craft well enough to evaluate AI output rather than just accept it. Have a clear enough positioning that “AI can do that too” does not hollow out their value proposition. Integrate AI into their workflow in a way that compounds — context that builds, reasoning that sharpens, output that reflects accumulated knowledge rather than isolated transactions.
None of that requires fear. It requires paying attention and doing the unglamorous work of being precise about what you actually do and why it matters.
The noise will not stop. But it does not have to be your signal.
The positioning question — what you actually do, who you do it for, and why it’s hard to replace — is worth its own treatment. I wrote about it from a strategic angle over on andredaus.com, for those who want to go deeper on that side of the argument: When Smart People Ask the Wrong Questions