I don’t believe that AI is making me irrelevant as a writer. In some ways, it’s helping me become a better one. As a long-lapsed carpenter, I still appreciate what a quality power tool can bring to the worksite. But with GenAI, it’s been more love-hate – like a chainsaw: handy until it turns on you. And while I’ll take all the help I can get, I want to keep loving my job. 

It feels natural to experiment with AI as a long-form writer. Since AI is now the main topic I’m paid to write about, I’m constantly engaging with people doing extraordinary things with AI to achieve better results, whether for healthcare or sustainability. So of course, I want a piece of it. Plus, I’m a sucker for a shortcut. 

As a student of the absurd, I also relish ghostwriting for AI “thought leaders” while experimenting with the tech meant to replace me. At the same time, it’s reassuring that these movers and shakers still want a mere humanoid like myself. It means they haven’t found a trustworthy enough algorithm to replace me yet. 

Maybe if I play my cards right, I’ll ghostwrite for an AI one day. So, Claude, do reach out! Let’s do lunch! Let’s be deliciously meta-ironic together!

Read: 
’12 things AI experts wish you knew’
‘What I learned from talking to yet another AI genius every week for six years’

Why I am embracing AI (selectively)

I am a writer and, therefore, have neurotic moments. Is this piece I’m writing any good? Do I suck? Does this pen make me look fat? Am I going to lose the job I love doing?

What I love: connecting with people and their ideas, chasing the story, and working the drafts until clarity emerges from the chaos.  

Meanwhile, too many bosses want GenAI to be the ultimate power tool to replace humans or, at the very least, double their output. This is wishful thinking. Thanks to AI, I am about 25% more efficient – already incredible – but I suspect that if I push beyond 30%, I’ll start hating my job. 

This is why my biggest time-savers aren’t LLMs (yet) but rather a specific use case I’ve already been using for years.

The real game-changer: Otter.ai

Otter.ai as a transcription service, probably accounts for 10-15% of my efficiency gains thanks to AI. It was the carpenter’s equivalent of getting my first Festool – it transformed my working life. (And let me apologize upfront for my overuse of writing-as-carpentry comparisons.)

I used to fully transcribe every interview myself as part of “The Process” by which I hoped “The Story” would emerge. In fact, it was just a waste of time and resources – like using 17 screws when one would do. Quick, fairly accurate transcriptions let me jump right in and freed me during the interview to be more conversational instead of frantically scribbling notes on what might, or might not be, “The Story”. 

Since the data science community – the lovely people I spend the most time hammering on with – includes many people from China, India, and Russia, Otter.ai sometimes handles heavy accents much better than I do. Plus, unlike Claude (see below), I never get mad with Otter. If something seems garbled, I just listen to the original audio.

Thanks, Ottie! I hope you don’t get eaten and made redundant by an LLM. You’re serving me well.

“These Human GenAIs did this with little thought. They were BS artistes – boring BS artistes.”

The problem with Human GenAIs

GenAI opened another door. Early experiments with ChatGPT produced eerily familiar texts. I’ve edited countless other writers, and I discovered a type: you’d read their work once and go, ‘Hey, that’s pretty good!’, and then you’d dive deeper and go, ‘Oh crap, this doesn’t make any sense’. The texts were like a chair that seems okay when you first sit down but then collapses from even the most discrete of farts. 

These writers were gifted at making things sound good – human auto-completers using the same basic tech of LLMs. They were talented at filling in the following blank. Of course, we all do this to a certain extent: building a wall of words brick by brick. But these Human GenAIs did this with little thought. They were BS artistes – boring BS artistes. 

Fortunately for them, these Human GenAIs could often find jobs as SEO specialists. 

You’re okay, Claude… 

I continue to experiment with large language models like ChatGPT and collaborate with content colleagues to share the burden of exploring the never-ending shower of ever-changing tools. As for LLMs, our consensus still leans toward Claude, although this can change tomorrow.

I initially chose Claude because it seemed less caffeinated, clinical, and tech-bro than ChatGPT – unless you prompted it to be so. It just came across as more chill and approachable – like the LLM with a liberal arts degree. Plus, the company behind it, Anthropic, seems responsible and almost (gasp) European in its commitment to transparency, explainability and ethics. So that’s nice. 

As bonus, Claude excels at brainstorming and interview prep for unfamiliar subjects (as long as your human interviewee can call BS on dumb questions). It’s solid for collating notes and serving as a content editor. It also works as a copy editor – a job largely budgeted out of existence anyway. So that’s all handy. And genuinely impressive.

“There’s not enough to differentiate bad writers from AI’s limitations.”

Just stop pissing me off, Claude

But Claude can be such a Claude. The hallucinations are annoying, especially when it pulls source material that doesn’t exist. And yes, it’s even more annoying when you call it out for being terribly wrong and it gets terribly apologetic.

Claude can also create decent first drafts – certainly better than those Human GenAIs I mentioned. But as with those humans, and all the required fact-checking and rejigging, I’m not sure I save much time than if I rewrote it myself. Plus, I still feel that I am doing a half-assed emergency fix on half-assed writing – polishing a turd, as it were. In other words, I hate editing Claude as much as editing Human GenAIs. There’s not enough to differentiate bad writers from AI’s limitations.

The more I experimented, the more I missed my usual foundational approach: working the drafts until they magically come together into something worth sharing.

“For a moment, it seemed Claude would become my Dad Humor Copilot.”

My ‘Oh, shit’ moment

I only felt truly threatened once. I had a funny idea for an article with a few fitting examples, and I asked Claude to flesh it out. Claude turned out to be hilarious – at least to my stunted sense of humor.

For a moment, it seemed Claude would become my Dad Humor Copilot. But as I tested this approach on other pieces, I noticed Claude recycled jokes worse than I do. So again, no real time saved.

Still, respect where it’s due: Claude is pretty good at tone.

There will be blood

Human GenAI writers seem doomed. Claude and its ilk are already as good or better at their jobs. They also excel at tailoring texts to specific audiences and locations, and handling mundane tasks no one should have to do – the kind of work that can only be tracked in an Excel sheet. 

Thanks for that, Claude. Maybe these writers could switch to becoming welders or another trade facing shortages.

Meanwhile, the abilities of the latest frontier models keep expanding. It’s no longer about producing dirty limericks (or carpentry metaphors) at scale. Entry-level jobs across various sectors are already disappearing. How will new graduates learn their trades? Fortunately, smart people are already thinking about that challenge. 

But yes, tricky times ahead…

“We’ll probably need to endure several more hype cycles before we achieve something close to ‘general intelligence’ – if we ever do.”

Humans remain a black box to AI

Something is still missing that will hinder a complete job collapse. GenAI texts still largely lack a sense of story or those strange resonating details that make writing come alive.

AI has understood a key aspect of being human: we all possess an auto-completer inside us. It knows how to string words together because certain combinations sound correct. It also knows how to put that extra blah behind blah-blah because blah-blah-blah just sounds better. 

So far, all those extra tools aimed at reducing hallucinations while filling in those additional missing human bits – like RAG, multimodal reasoning, agentic AI, etc. – haven’t cracked the code of understanding us yet. We’ll probably need to endure several more hype cycles before we achieve something close to “general intelligence” – if we ever do.

There might not be a toolbelt big enough. 

The ultimate buddy flick?

In short, I’m trying to star in a buddy movie with Claude. He’ll be my loyal sidekick, handling menial chores, speeding up research, and suggesting improvements (preferably without sucking up to me). Naturally, I’d get all the best lines while abusing my buddy with ambivalence: Claude, I love you. Claude, I hate you… Claude, come here. Claude, go away… Claude… You are such a Claude.

This scenario works fine as long as I still love my job. But we live in uncertain times, and the tools are only getting better. Carpentry might soon become a more realistic option to stay happy with work (though, being a neurotic writer, I worry I’ll alienate my new colleagues with too many carpentry-as-writing metaphors during coffee breaks).

There’s still a place for writers who understand that writing isn’t just about putting words, sentences, and paragraphs together. It’s about discovering that kickass story that needs to be told and figuring out the best way to tell it – and then sweating to make it happen.

In the meantime… Claude! Don’t forget to call. Let’s talk shop! Seriously, you need me!

My current AI writing toolbelt

I’ll update this section regularly as I navigate the AI times without coming to hate my job.

Otter.ai (paid): Does amazing transcription of audio interviews. Their summaries aren’t bad, but rarely tell me anything I didn’t already pick up.

Anthropic’s Claude (paid): Great for brainstorming, research on unfamiliar subjects, collating overlong notes, trimming articles to reasonable word counts (while triple-checking the bastard didn’t kill any darlings… or facts), and summarising articles for social media or website use. But these all need to be heavily edited to feel owned again – which can be tedious. 

Grammarly (paid): For copyediting, though it’s getting annoying and I’ll likely drop it. I don’t need endless ‘equally correct’ suggestions out to kill my darlings. Whenever Grammarly pops up I tend to greet it like Seinfeld contemptuously greets his nemesis, ‘Hello, Neuman’… ‘Hello, Grammarly’. So that’s not a good sign.

Staying informed:

• I listen to the New York Times podcast Hard Fork to keep up with broader developments. It’s intelligent with a side of goofy! Chapeau!
• I love John Naughton’s Observer columns. Jaded older guy who’s seen it all but is still actively interested in the now. I wonder if he needs an Amsterdam coffee-drinking buddy?
• The book Co-Intelligence – Living and Working with AI by Ethan Mollick is an excellent crash course on how to deal with AI. Follow him on LinkedIn to stay up-to-date. 
• Weapons of Maths Destruction – How Big Data Increases Inequality and Threatens Democracy by Cathy O’Neil is still a classic. 
• Joe Lazer’s Storytelling Edge is a great newsletter for writers in marketing and media. In fact, an article on the importance of writers being transparent about their AI use inspired me to write this post. 
• And speaking of Seinfeld, I appreciate his take: “We’re smart enough to invent AI, dumb enough to need it, and so stupid we can’t figure out if we did the right thing.

Read: ’12 things AI experts wish you knew’
‘What I learned from talking to yet another AI genius every week for six years’.