At this festive time, looking back with nostalgia on the golden age of thinking for oneself.
I have to confess, I found it quite a struggle to begin writing this article. But not for what one might consider the ‘traditional’ challenge of a blank page. No, this was something more surprising, and furthermore something with a hint of the withdrawal symptoms of an addict. I had to battle, with no small effort, the compulsion to get AI to write it for me.
At Katatomic we’ve been proud early adopters of AI in our business. In the summer of 2022 we had begun to experiment with an early iteration of Midjourney to supplement client-supplied images, and then soon after when ChatGPT had its breakthrough moment in November of that same year we had within weeks begun to incorporate the power of an LLM into our workflows and were loudly extolling the virtues to our clients.
Today, the benefits we derive on a daily basis have increased exponentially, and there’s a mixture of excitement, perhaps tinged with a hint of disquiet, as to where the future is leading.
But it’s a double-edged sword. There’s sometimes a palpable desire to abdicate thought to LLMs.
The temptation is entirely understandable. Why wrestle with a tricky paragraph when you can describe what you want and have something serviceable appear in seconds? Why puzzle over how to structure an argument when your ‘best friend’ (an initially humorous nomenclature I use which is now becoming less funny) will arrange your scattered thoughts into neat, logical order? The friction of thinking, that uncomfortable mental effort that accompanies creative work can simply be, well, bypassed.
And therein lies the problem.
The muscle that atrophies
There’s a reason we don’t outsource physical exercise to someone else. (Would that I could!) The point isn’t merely the output, a certain number of steps taken, weights lifted, but what the process does to us. The same principle applies to thought. When we consistently hand over the hard work of thinking, we’re not just saving time, we’re allowing a cognitive muscle to weaken.
I noticed this in myself a few months into heavy AI usage. Tasks that once came naturally, like drafting an email, structuring a proposal, suddenly felt more of a chore. Not because they’d become objectively harder, but because I’d become accustomed to not doing them. The path of least resistance had reshaped my habits.
Where AI genuinely excels
None of this is to suggest we should abandon these tools. That would be rather like refusing to use a calculator because mental arithmetic builds character. The benefits are real and substantial.
In our development work, AI has become an invaluable collaborator. When you’re deep in a complex WordPress plugin, wrestling with some obstinate piece of logic that refuses to behave, having an assistant that can examine your code and spot the missing semicolon, or the flawed assumption three functions back, is genuinely transformative.
It’s also remarkably useful as a reference tool. The web development landscape is vast, and no one holds every technique in their head. When we need to implement something we haven’t touched in a while, a particular API integration, an obscure WordPress hook, a specific JavaScript approach, AI can rapidly surface the relevant methods and syntax.
The question isn’t whether to use AI, but how.
Adopt the role of an editor
The distinction I’ve come to find useful is between generation and judgment. AI is excellent at generating code, suggesting approaches, offering solutions, but the judgment of whether that solution is actually appropriate for the context? That remains stubbornly human.
We know, from twenty-five years of building websites, what constitutes a robust solution versus a fragile one. We understand the implications of a particular approach for maintainability, performance, and security. AI can suggest code that technically works but creates problems down the line. Over-engineered solutions, unnecessary dependencies, and approaches that don’t work alongside the rest of the application.
The most effective approach treats AI as an assistant, never an authority. Let it generate options but decide yourself which to pursue. Let it suggest a debugging approach, but you verify it actually addresses the root cause.
This editorial role isn’t passive. It requires genuine expertise to recognise when something is merely functional versus properly sound. These are precisely the skills that atrophy when we accept whatever the machine produces without scrutiny.
The client question
We’re occasionally asked whether we use AI in our work. The answer is firmly yes, and extensively, but not as a replacement for expertise. The experience we’ve accumulated means we know what questions to ask, what pitfalls to avoid, the subtle differences between something that merely works and something that works well.
AI amplifies our agency expertise. Without it, AI is just a very fast way to produce mediocrity.
Maintaining the balance
So how do we resist the pull toward intellectual abdication while still benefiting from these remarkable tools?
Firstly, by being deliberate. Recognise when you’re reaching for AI out of genuine efficiency versus simple avoidance of mental effort. Sometimes the struggle is the point.
Secondly, by staying critical. Don’t accept the first output. Push back, refine, demand better. Treat AI suggestions as input to your thinking, not a substitute for it.
And finally, by continuing to do hard things. Write without assistance sometimes. Solve problems the long way occasionally.
If you want a glimpse of where the wholesale abdication of thought might lead in the long run, try reading The Time Machine by H.G. Wells. Neither you nor I want to end up like the Eloi.
Of course, if that reference means nothing to you, you can always ask AI.