The reassurance of the hand

Artisans have always commanded premiums for skills honed over decades, but before the industrial revolution, the label “hand-made” was redundant. From pots to palaces, everything was the result of manual labour.

For a time after the rise of the machines, “hand-made” might actually have signaled inferiority. A garment not produced with the efficient uniformity of a factory was likely to bear the signs of rushed or sloppy work. Critics like John Ruskin and William Morris, who bemoaned the death of craftsmanship, often ignored a simple reality: machine-made goods were affordable for the masses, not just the elite.

Only recently has “hand-made” regained its status as a mark of careful attention. Today, that same shift is hitting white-collar work.

Billy v Queen Anne

First, I think it’s important to acknowledge that when it comes to corporate content, a vast amount of AI work is perfectly serviceable. It serves the same purpose as an IKEA wardrobe: functional, harmless and stylish enough. That said, it’s unlikely to be mistaken for a Queen Anne armoire. In fact, it looks out of place in many environments and, like a Billy bookcase in a Hampton Court bedroom, its use can even cause offense.

The difficulty is that the "flatpack" nature of AI-generated prose is harder to spot than a laminated woodchip shelf. Readers feel the difference in style, rhetoric and metaphor even if they can't articulate exactly what it is, or why the sparkless sentences they’ve just laboured through haven’t ignited the fire of their attention.

Because they are still learning to identify the veneer of machine-generated content, people often look for what they’ve come to believe are crude "tells" – specific AI-favoured words and punctuation marks in articles, or weird fingers in images. But these supposedly surefire indicators are often far from reliable.

Stand it up, or stand down

In fact, when it comes to AI, the comparison to the Industrial Revolution is reversed. The Spinning Jenny made linen more consistent; GenAI, being probabilistic rather than deterministic, often makes content less reliable. In this era, the machine-made output may be more likely to contain flaws.

I was reminded of this recently when negotiating a fee to repeat a project we’d successfully completed for a client last year for US$20,000. Bearing in mind rising costs and the additional resources experience had taught us would be needed, I proposed a 30% increase; the client, understandably keen to keep budgets down, countered with 0%. Gemini then cheerily suggested an email response for me that an acceptable compromise was $25,000, since that was “only a 12.5% increase from 2025”.

Thankfully Gemini didn’t try to be helpful by sending the e-mail too; given the financial expertise of many of the companies we work with, we want our clients to remain confident in our ability to grasp simple maths.

Unlike this confidently asserted innumeracy many errors, much like knots in IKEA pine slats, won’t matter. But where they do, they can deal substantial damage and reinforce the truth that there is no alternative to human expertise. You cannot trust an AI-generated work to withstand rigorous scrutiny from those with deep knowledge of the subject matter. If you omit human-in-the-loop quality control, you will eventually pay the reputational cost.

Ultimately, too, humans are required for accountability. I am reminded of an IBM training manual from the late 1970s that warned: “A computer can never be held accountable. Therefore a computer must never make a management decision.” Someone, at some point, must stand behind the work – and you can only do that if you have done the work necessary to "stand it up".