When AI tooling became easily and cheaply available to the general public I jumped on it.
For me, it all started with Midjourney. For years I kept a paid account, and I loved playing with the image generator. I even tried to sell some of my designs as drop-shipped t-shirts, but that failed miserably. Turned out people who like buying funny t-shirts like to think it’s hand-made art on them. I’m not going to digress into why I think that’s a somewhat untenable position to take, but I do want to mention that I should have taken note of something I totally missed.
And that is at the bottom of a lot of the backlash we see currently around the AI discussions. People don’t like to think that something they value is mass produced and machine made. In some respects this has become the norm, and it’s widely accepted – but only to a point.
Take fast fashion: the vast majority of people will buy their clothes in one of the high street shops, and are more or less aware of the horrible conditions their clothes are manufactured under. A certain percentage of people will pay slightly more money for brands who assure them their workers are treated better. A very select few will be able to afford tailored clothes, and some will learn to sew to make their own (that would be me). People would still rebel if they thought their £3 Primark t-shirt had been designed by a robot (but happily accept an AI stylist).
I don’t want to repeat my whole argument about cheap production here, although I stand by it. I’ll just recap the focal point of that argument: once the production of a thing has become cheap and so easy that (cheap) low-skilled workers can do it, there is no going back. Once you can make a shoe with a machine, the skilled shoemaker becomes the aberration. Do you own a pair of handmade Italian shoes? I don’t, I couldn’t afford them. But I can afford reasonably well-made mass-produced shoes, so my toes stay toasty in the harsh Yorkshire winters.
There are, of course, pros and cons to this. And it’s more or less true for different areas of our work and life.
Anyways, back to AI usage.
After loving Midjourney for a good while I stopped using it. Not so much because I didn’t like it anymore, but because other tools stepped in that were more interesting to me. My main working tool is language, and the words I use.
So I got sucked into text creation with LLMs. First it was ChatGPT, then I tried Gemini (which was really bad at writing at that time), then Claude. For a while there was a bit of fluctuation every time the frontier models were updated. Sometimes ChatGPT was better at sounding vaguely human, then Claude, then back. Then I discovered working with Claude Code in my IDE, and I’ve never looked back.
I started setting up guardrails, and rules files, and skills and agents and a wiki. And all of that made Claude sound less like Claude and a little more like me. But it still wasn’t me, and eventually it all came tumbling down.
Over summer, I gradually fell out of love with my ways of working. I still think it worked reasonably well, and I managed to remove a lot of the “AI tells” from what I call my utility texts. Because if you’re a one-person business, and you’ve got to do the marketing, the teaching, the creating teaching materials, applying for workshops and conferences, the travelling and whatever else your business needs it’s A LOT, and one is happy for any help one can get. Utility texts, for me, were texts that had information value for you, my dear readers, but weren’t necessarily pieces dripping in my heart blood and the sweat it had cost me to pen them. Texts where clarity is more important than character, texts where the message is clear and is generally centred around getting knowledge and information out of my brain into a shareable form.
I’ll briefly describe my workflow for those. I’d dictate a thought or an idea, then I’d have that transcribed to text and give it to my Claude setup. There a skill would take over: find the actual idea, form it into an argument, decide on a structure to present it, and flesh it out. A lot of this happened in a question-and-answer between me and Claude. Eventually it would present me with a draft, I’d go through it with a fine-toothed comb and fix the bits I didn’t like. The whole process would take anything from 30 minutes to an hour. The core of it was still me, the language was shaped on my previous writing, and I’d have full control over content and ultimate form.
And yet, at some point during the last year, I started to dislike these pieces. Despite trying to make them “me” as much as I could, they … just weren’t.
And this same feeling of alienation applied to other things I did. My presentations started to feel less exciting to myself, my workshops felt a little uninspired.
Do I regret my overuse of AI? Only insofar as I realise it put quite a burden on you, my dear readers. For that, I apologise. For me it was a wild ride, and one I deeply enjoyed because I’m the kind of person who loves the steep bit of the learning curve. The bit where everything you do that works feels like a milestone achievement, and I get a bit bored once I’m past that and every improvement is very incremental.
I also almost ruined my joy in what I do, and that’s a bit of a shame. But I’m determined to claw it back. I’ll continue to use AI tools for many of the mechanical aspects – I’ve built an agent that takes a written blog post and puts it on my website, for example. I don’t enjoy fighting WordPress, and it genuinely saves me time, so why would I stop using it?
And I don’t want to manually search for that email attachment I know somebody sent me ever again. Gemini is actually really, really good at finding all relevant emails for a given topic, summarising what’s in there, what attachments I have and what action items are outstanding. It’s life-changing.
But when it comes to my writing, my presentations and my workshops – I’ll be going back to being a lot more hands-on. I actually enjoy writing, so why did I ever think removing that from my workflow was going to work?
And maybe because I’ve seen it so much, there’s the other side to this: I have developed a very fine nose for AI-generated text. As it turns out, when it comes to human writing I, too, prefer the handmade Italian shoe version of it. I want to read the unique structure of somebody’s thoughts, and the words they chose to express them.
Even with the good AI-generated texts, the second I get a whiff of them, I stop engaging. My brain just switches off, and I move on. And I realise that you might feel the same, and I can’t blame you for that.
That’s how my AI use has gone so far. What I’ve learned from it deserves a post of its own, so I’ll save that for next time.
I’d love to hear how your AI journey is going. What do you enjoy, what do you hate – and are there maybe parts that you thought were marvellous but have grown tired of?