The newsletter I deleted

When iteration becomes abdication: confessions from the AI learning curve.

I’ve been working on my newsletter for this week for several days now. It was formed from a place of deep thinking, curiosity about what “all this AI means” for me, for how and what I learn. I spent hours writing out all my thoughts. I’ve been concerned about protecting the spaces where my brain still has to work, where I still have to wrestle, where learning still costs me something.

I explored the concept of rented knowledge versus owned knowledge, the potential loss of critical thinking abilities. I wondered: if we only rent knowledge, do we lose the ability to think deeply when offline? Do we stop making unexpected connections because nothing’s stored to connect? I dove deep into the importance of creating areas of friction so that we keep learning and stay engaged. In a world where everyone can rent infinite knowledge, what makes someone actually expert? I learned about the Google Effect and digital amnesia. Being able to create more doesn’t mean I understand more or produce quality; having answers doesn’t mean I’ve done the thinking. “Desirable difficulty” was discussed – the importance of having the kind of challenge that actually strengthens learning. When you have to work to remember something, figure something out, solve a problem, you’re literally building and strengthening neural connections. Your brain encodes the knowledge through the struggle. I explored how to decide what knowledge to own.

It was all very interesting and an important conversation.

This is not that newsletter.

I spent hours writing down my thoughts last week. There were a lot of them floating around my head and they had been driving me crazy as I was struggling to see how they all connected. I knew there was a theme and my brain was focused on teasing out a meaning, but it just wasn’t clear. So, I wrote. I got it all out. I wrote until all my floating thoughts were in black and white on a Word document in front of me. That felt good. The sitting with discomfort and uncertainty was important – this was a turning towards friction.

My next step was to add this document into Claude and ask for a critical review, to use the tool to explore what the common threads were. It’s absolutely fascinating to ask AI to be a mirror, and Claude nailed it. I asked AI to reflect back to me what it saw. Did I make sense? Were my thoughts connected and how? What was I missing? I had my “ah ha!” moment after so much time spent mulling over what it all could possibly mean. That felt marvellous. What I was getting back from the AI tool resonated and I felt like I’d found that golden thread.

I then used Claude as my collaborator to iterate and polish up my work to a sparkly impressive shine. This was my second layer of wrestling with content until it felt complete – all very in keeping with what I was exploring.

The newsletter was finished late last week. It felt good. Until it didn’t.

Something just started to feel wrong with it. I had the luxury of time, so rather than pressuring myself, I read through my work again a few times over the following days. I made a few edits, I tweaked a few things. It still just felt “off”, like I had strayed away from where I started but I wasn’t sure how.

There was a good dose of telling myself, “Dallas, you’re just overthinking this!” I am a “somewhat reformed” perfectionist and I do have a tendency towards being rather OCD about my work, particularly when it’s important to me and I want it to be correct. I was aware that this might be the case, but I’ve also learned a lot in recent years about leaning into these thoughts and trying to hear what they are actually telling me underneath it all.

Yesterday, I began a serious re-edit. I asked Claude to critically review the work again, to systematically check for errors, correct any assumptions stated as fact. I explored what might be missing from my argument, points of view I had not considered. We were now hitting a serious contender for “most polished newsletter Substack has ever received”.

Then I realised the problem.

Everything was going on a good trajectory, but it had stopped feeling like me. In the iterative process, I had lost my voice and the many lived experiences which illustrated my work initially had given way to science. Science is great, I love science, but I’m not an expert in any way. Claude had provided a lot of excellent research which backed up my conversation, but this all needed to be verified. It was clear I had now unintentionally waded into territory where I was presenting a case that was heavily dependent on Claude to do the heavy lifting. That made me uncomfortable.

I wanted the thoughts to be mine. I wanted to own them, to have contended with them. I had drafted the ideas, the memories, the questions – they came from me sitting with the disquiet, an unknowing. Now, in this polished version, I was presenting myself as something of a cognitive scientist, a pseudo subject matter expert, and I wasn’t okay with that.

As I started the process of transferring my work over to Substack and worked on the formatting, statements started standing out to me that I realised I needed to check, yet again. As I ran them through Claude, each time without fail it would say, “oh, this is not entirely correct”. Even statements on what AI could and couldn’t do had still been framed incorrectly, by the AI, over numerous iterations and prompts to check for error, hallucinations and assumptions. The new revisions felt better, less absolute, more wriggle room.

AI had not just helped answer my questions, it made me sound expert. This is illusory expertise. In reality, I had no idea if I was presenting true fact, pure fiction or something murky in-between. I lost perspective. A risk had been introduced to my newsletter that I could not accept.

I asked a question in my newsletter: “In a world where everyone can rent infinite knowledge, what makes someone actually expert? I’m wrestling with this personally. How do I stand out as someone with deep ownership versus someone just selling knowledge I’m renting on the fly?” This seemed to be a case in point – I was renting knowledge. I didn’t know what I didn’t know. I also asked, “perhaps the only thing that truly makes you valuable is what you’ve taken the time to understand deeply?” I had to sit with these questions and evaluate what I had created.

I finished formatting my newsletter yesterday and decided to let it sit overnight to see where it landed in a morning read through and perhaps run what I had created in Claude through ChatGPT for a final final review. I think we’d hit pure stubbornness by this point!

I didn’t do that. I deleted my newsletter first thing this morning.

It was simply not something I could put out into the world. Seasoned experts are probably reading this thinking, “Oh sweetheart, that clue way back there was your sign! Why didn’t you listen earlier?” You know, next time I probably will (a fraction earlier at least)! I had started out so excited by the topic of my newsletter, I had felt creative, my thoughts found clarity and I was determined to create something magnificent. Until I went off course. Until I stopped listening to myself and allowed AI to dictate the path. In reflection, I should have stayed with my thoughts and just presented the threads. I didn’t actually need to provide any answers about what it all meant, but I allowed myself through repeated iterations to be carried away in a rip tide of my own making.

It all reflects a theme that started this whole journey – how do I learn to partner with AI while not letting it do everything for me? It’s strategic offloading versus thoughtless dependency. The question isn’t whether AI can give me answers; it’s whether I’m still doing the cognitive work that makes those answers mean something. I need to run towards friction where it matters, and before I take off running, I need to have clearly identified what direction I need to run and why.

I sit with more questions now…

What do I need to own deeply enough that I can use AI effectively rather than be used by it?

Where am I choosing friction intentionally and where am I just avoiding necessary evolution?

Am I building the metacognitive muscles that let me evaluate what AI tells me?

What makes me irreplaceable in my specific context?

The rent-versus-own framework in practice is that I own the thinking, the wrestling, the judgement. From AI, I’m renting the reflection, the gap-spotting, the fact-checking support (but clearly not all of it based on recent experience!).

The order matters. If I let AI do the thinking first, I’m not learning – I’m just editing someone else’s work. However, if I think first and then use AI to pressure-test and strengthen my ideas, I’m building my own cognitive pathways whilst getting smarter faster. The human in the loop principle must always be in place. It begins and ends with me.

My first newsletter could have easily been published. I might have even received a few likes, maybe a nice supportive comment or two, but there would have also been people who read it and “walked away”. Why, when I am working so hard to build credibility, would I undermine it with what could be viewed as very polished AI rubbish?

I actually feel good right now about starting again. It feels authentic to the voice inside that was cautioning me, the one that pushed for so many revisions and still wasn’t comfortable. That inner voice kept illusory expertise from becoming fake expertise. I’m acutely aware that I don’t know what I don’t know. I won’t always know enough to recognise when what AI gives me is wrong or when I’ve gone too far. I need to learn how to put robust guard rails in place and I need to be vigilant with this.

This is my journey – learning to fail fast, pivot and adjust, and keep moving forward. I’m excited about the future I can have using AI, but I’ve learned that as well as the need for friction, I also must keep my voice. Not just that my work here sounds like me, but that it is me.

Stay curious, stay critical, stay you.