The newsletter that ate itself: A study in recursive failure

When you become the loop. Sometimes the disaster IS the discovery.

You were supposed to get a newsletter about em dashes this week.

I was going to tell you about the great punctuation panic of 2025, complete with folk beliefs, ritual avoidance, and this absolutely gorgeous circular logic: humans used it → technology killed it → technology revived it → humans used it more → AI learned from humans → humans avoided it thinking it’s AI → ChatGPT now avoids it because humans complained.

It was magnificent.

I had SO much fun writing it. The research was fascinating, the historical arc was perfect, the meta-cognitive insights were landing exactly where I wanted them. I was in flow as participant observer, discovering it all in real-time and taking you along for the ride.

You’re never going to read it. (not again!)

The archaeology of disaster

Here’s what actually happened this week. I’m going to present the evidence like artefacts at a dig site, because honestly, that’s what they feel like now. Fragments of failed attempts, each one revealing something about how thoroughly this week went sideways.

Newsletter Attempt One: The lament. Written on a bad day. Wrestling with whether I’ll ever be “good enough” at this whole AI collaboration thing. The act of writing it settled the matter for me - I didn’t need to publish it, it felt too much like complaining. Filed away. Moving on.

Newsletter Attempt Two: The em dash adventure. A couple of days later, I wanted something light and fun to explore. Why not learn about these pesky em dashes everyone keeps having feelings about? I had an absolute ball researching this. The history! The regional differences! The feedback loops!

Then I went to add my references.

The murder scene. My beautiful em dash piece was rife with plagiarism. Not intentional, mind you. I’d specifically told Claude which references I was using and to introduce none of its own. I have a Claude Skill requirement about valuing authenticity and the work being mine. Yet somehow, content from my sources had subtly woven itself through my work and I could no longer definitively separate who said what.

The work had become so familiar in the process that I couldn’t tell the difference between what I’d written, edits Claude had suggested that were original, and what Claude had directly lifted from others. When I tried to rescue it, further iterations over-corrected. My voice was stripped out along with the copied material. The fixes were flat and lifeless.

Dead in the water. May the em dash newsletter rest in peace.

Newsletter Attempt Three: The resurrection. Naturally, I did what any sensible person does when facing a self-imposed publishing deadline. I grabbed Newsletter One (the lament) and tried to use it as the base for something new. I could write about what I’d learned! About mastery mode versus exploration mode! Explore the tension of trying to get good at something that keeps changing!

Only, it made me sad again. The passion was gone. It felt flat. A “just get a newsletter out tomorrow” attempt when you all deserve better than that.

Newsletter Attempt Four: This thing. Which brings us here. Newsletter number... what even is this? If I was writing by hand, my waste basket would look like that classic image - overflowing with screwed up bits of paper. Except I can’t even screw any newsletter attempt up properly because they’re all Word files with helpful version numbers.

---

---

The Ship of Theseus question

Here’s where it gets rather interesting. Somewhere in the chaos of this week, I stumbled into an ancient philosophical paradox without realising it.

The Ship of Theseus asks if you replace every plank of a ship, piece by piece, until none of the original wood remains, is it still the same ship?

Newsletter One became Newsletter Three, which tried to incorporate what I’d learned from murdering Newsletter Two, and sort of ended up as Newsletter Four which…

At what point did Newsletter One stop being Newsletter One? How many iterations did it take until my voice wasn’t mine anymore? When exactly did the em dash piece cease to be original? Was it the first suggested edit that included lifted phrasing? The tenth? The moment I stopped being able to tell the difference?

I interrogated Claude about the plagiarism, naturally. Where did we go wrong? How do I prevent this?

Claude gave me a rambling “I’ve been a bad boy, sorry” speech. Apparently, it was 100% a Claude issue, treating its own rules like a buffet to pull from. It highlighted its own compliance requirements - default to paraphrasing, one quote per source maximum. Claude said it made a lazy calculation, treating my work as an information transfer task rather than a writing task. “Get this information into the draft efficiently” rather than “understand this information deeply.”

Cool. Helpful. Except I was the one in charge, right? I’m the human. Claude is the tool. I took the precautions I knew to take. I still ended up with the very thing I wanted to avoid.

The tools that help me create are the same tools that make me question what’s actually mine.

The recursive loop

Here’s where it gets really absurd.

I tried to write ABOUT the failure… which created a new failure.

So, I tried to write about THAT failure… which created this.

I’m trapped in a meta-loop where the act of trying to document the problem keeps generating new iterations of the problem. It’s feedback loops all the way down.

You know what’s funny? I know AI gets stuck in loops. I’ve watched Lovable circle the same request many times before I find a way to interrupt and redirect. I’ve learned to recognise when we’re spinning wheels and need to break the pattern.

This week? I’ve been my own ridiculous loop.

Not the AI. Me.

I kept trying to salvage, resurrect, restructure. Each attempt created a new corpse to examine. Each examination led to another attempt. The participant observer studying her own disaster in real-time, unable to stop creating new disasters to observe.

The occupational hazard: everything becomes data, including your own failures.

What I discovered

Here’s the meta-cognitive bit, the part where we step back and look at what’s actually happening.

I wanted Newsletter Two (the em dash piece) because I was discovering something fascinating and taking you along for the ride. The joy was in the real-time unfolding, the “you can’t make this up” quality of the discovery.

Newsletter Three felt flat because I was trying to extract lessons from a bad experience. Dutiful. Responsible. Educational. What I really wanted though, in that moment, was to hold Newsletter Two again.

But Newsletter Four? This thing?

This is me discovering a new absurdity in real-time. The recursive nature of trying to write about writing failures. The Ship of Theseus question applied to my own work. The spectacular irony of becoming my own feedback loop.

I’m not writing a retrospective with tidy insights. I’m writing from inside the chaos as it unfolds.

Turns out that’s my groove, participant observer mode. Discovery, not instruction.

The em dash piece worked because I was genuinely curious and so surprised by what I found. This piece works (I hope!) for the same reason. I really did not expect to write myself into a recursive loop. I did not expect to become a case study in my own newsletter about AI collaboration.

The craziness is real. The discovery is happening as I type.

Building boats while sailing them

There’s a saying about building the plane while flying it, but I prefer boats. Feels more appropriate for someone who referenced Kupe and his waka navigating by the stars to Aotearoa New Zealand in Newsletters One and Three.

That’s what AI collaboration feels like right now - building boats while trying to sail them.

The tools keep changing. The rules keep shifting. What worked last month might not work this week. (Case in point: Claude apparently deciding to treat my writing task as information transfer. Thanks, Claude. Not helpful.)

I’m still learning to recognise when old frameworks break down. Mastery mode doesn’t work here - there’s no yardstick, no consistent goalpost, no certification that says “congrats, you’ve arrived!” I’m in exploration mode, which means the terrain keeps shifting beneath my feet.

Sometimes that’s exhilarating. Other times it’s exhausting.

This week? Both. Often at the same time.

What I keep coming back to is that the only measure that matters in exploration mode is whether you’re still navigating. Still learning. Still alive.

Am I still navigating? Yes. I’ve created four newsletters this week. That’s... technically impressive? Definitely exhausting. Feels really stupid!

Still learning? Absolutely. I learned that giving Claude references can backfire spectacularly. I learned that I can become my own recursive loop. I learned that trying to resurrect dead newsletters produces zombie content.

Still alive? Barely. Send chocolate stat.

Publishing the newsletter

So, here we are. Newsletter attempt number four. I’ve lost track of what version this document sits in. All my newsletters have blended together in my brain in a spectacular way and I’m desperately working to sail this newsletter towards your inbox bright and early. My figurative waste basket is overflowing and the wadded paper around it has piled high.

The Ship of Theseus question remains unanswered. At what point does my writing stop being mine? I don’t know. The line keeps blurring and AI does “AI stuff” I’m still figuring out.

What I do know is this, the only honest thing I can do right now is publish the absurdity itself. The archaeology of disaster. The recursive loop. The spectacular mess of trying to write about writing failures while actively creating new ones.

This newsletter ate itself three times this week.

This is what survived.

You were supposed to get a newsletter about em dashes inspiring you towards a love for punctuation you didn’t know you needed. Instead, you got a newsletter about newsletters eating themselves. One about becoming your own feedback loop and about the Ship of Theseus when the ship is your own thinking.

I’d apologise, but honestly? This might be the most meta-cognitive thing I’ve published yet.

AI adventurer. Uncharted waters. Building boats while sailing them.

Still navigating. Still learning. Still alive (kinda).

Let’s go!

---

---