The Day My Parrot Died

(not pining for the Fjords)

Q: What is a limiting belief?
A: Anything which limits the flow of magic in my life.

For several decades I held a belief that was unknown to me and which I never said out loud until yesterday:

I am a mimic, fluent enough to escape detection but lacking any understanding. When I am uncovered only bad things shall happen.

Now that I have decided to actually test its truth I can see that it has been lying there, like an ocean floor, and shaped so much of my life, silent and unseen. Now that it is seen I just cannot un-see it. So many mysterious acts in my past and present suddenly make so much sense. It feels as if a huge weight has been removed.

It is the belief underneath my endless consumption of tutorials, the reading about reading about writing, the never ending groundhog day of preparation to one day publish. I can talk about craft with anyone, tell you what makes a scene work, what makes an argument land, why one opening line breathes and another lies flat. Still the whisper stings: you do realise this is b/s right? You have never tested any of this. You describe the landscape without ever having walked it. You found a map and are making a living by reading from that map. You are clueless and if you ever step into the arena, everyone will see you for what you are. All hat - no cattle.

The cruel elegance of that belief is that it cannot be dis-proven from the sidelines. As long as I never published, it stayed exactly as true as it needed to be. It protected itself by preventing the only experiment that could kill it.

Comfortably numb.

Who knows why, but yesterday something broke open and before I knew what was going on I had committed to running an experiment. Or rather, I co-created a machine that made not running it impossible.

Sound complicated? Yup, it is. It also revealed a question that has stuck in my mind a long time: Can there be something that is actually too simple to understand? Once again, yup, there is. Simply dropping a belief provided the relief.

So here goes, this is the story of those thirty-odd hours — the trigger, what got built, what it costs, and the September 1 date that now exists in public because of it.

Get out of your own way

It started with a podcast transcript. Greg Isenberg's Startup Ideas Podcast, an episode with Allie K. Miller — she ran AI at scale at IBM and AWS, and she now runs her business with an AI workforce of thirty-four agents, led by an AI chief of staff called Simon (the episode is here; worth your hour).

Two things in it went through me like a skewer.

The first was her observation that she had been "functioning at the limit of my own imagination" — that every task her agents did was one she had thought to assign, which meant her business could never be smarter than her busiest day. Her fix was to stop managing agents and start enabling them: written goals, wide scope, and a three-word prompt — do smart things.

The reason I got skewered was because the same thing keeps happening to me - I have been working for years now attempting to get out of my own way. So in some way the phrasing in the episode created some kind of trapdoor I just fell through.

The second event was quieter. Everything in her system existed so that a human with finite hours could stop being the bottleneck in her own ambition. I read that and in some strange way I felt that silent unseen belief shift. You know those nature videos where you are watching an unremarkable sea-floor suddenly erupt into life and a strange creature emerges from perfect camouflage? Boom, that sudden shift in perception happened.

I went from unconscious incompetence into action pretty much instantly. All the dominos I had been stacking with tutorials and tools and AI chatting just lined up and off they fell, one after another.

So I handed the transcript to my own AI — I work with Claude Code, a coding agent that lives in a terminal on a Linux box in my office — and asked it a plain question: build me a plan to make the daily business of my publishing work run like this.

What got built - in a day

I want to name the parts, because the specifics are the useful bit and vagueness is how these stories become slop.

As an aside, this triggered my slop detector. It was agent generated, and I have no problem with that. I tend to free-write using an app called squibler where every word entered is erased if I pause for more than five seconds. Does this make every word I write a source of shame? Another point, I have found agentic writing to generate the occasional laugh-out-loud joke, and it would never occur to me that such a result was in any way tainted, because it makes me laugh. End of discussion. So it is for my regular writing relationship with agentic review. Is it any good? That is all that matters.

End of aside and back to the topic at hand. Over the course of approximately twenty-four hours, working alongside the agent, I (we?) put in place:

A five-charter agent team. Not thirty-four agents — five jobs, written down like job descriptions. The breakthrough here is that they are job descriptions, there are verifiable outcomes. Does the agent execute the job spec accurately and if not why not? First among equals: a chief of staff that whispers me toward my workplace and away from the morning coffee - a brief I can act upon, which stops my wandering attention cold and gets the gramophone needle in the groove (what moved, what stalled, what needs a decision); It is pure nectar for my addled attention span, because I no longer am tied to phone or keyboard, I just speak aloud or draw or cut and paste or whatever I fancy and this agent manages to harvest that multi-modal chaos and turn it into food for thought, quite literally, and auto-magically. It is hard to articulate how utterly liberating this is.

Ambient-AI, an always-on food-for-thought generator, customised for my peculiar mental dietary requirements.

In essence it is an intake pipeline that watches a folder where my voice memos land and transcribes them automatically; a content engine that turns finished writing into platform drafts; two watchdogs — one watching the niche, one watching the funnel; and a regular oil-check - a friction review that asks why I keep correcting the same mistakes and what context the team is missing.

A Seal Queue. I think of this like a form of immune system responder. Nothing (thus far) — no post, no email, no page — goes out unless I approve it on a board of draft cards. The agents have increased their reach but not their responsibilities. Allie Miller says it precisely in that episode: the tier of risk stays the same, only the width expands. I review everything, same as before. The difference is that the queue now fills itself based on the roughage I feed in via the input folder.

Come to think of it, a name change is required, lest someone think I am raising a seal colony. Perhaps the agent will read this during a proof-read and auto-magically rename it to 'The Event Horizon'.

The transcription pipeline, now with added guard protection. My working habit is dictation — I talk into my phone, or my DJI mic, while walking, and those memos are the basic nutrients for everything I write. The pipeline transcribes them locally (whisper.cpp, running on my own machine, nothing leaves this host under the stairs). And here is something I think about often: when 'we' tested the transcriber against my recordings, it made six errors, and two of them inverted the meaning of the sentence. "It's just physically not possible" came out as "physically not impossible." A machine that is fluent in English produced, with total confidence, the opposite of what I said. Hold that thought. It matters in a minute.

Eden. Oh what an appropriate name! The same day, I onboarded to Eden, Dan Koe's workspace for creators, and let it interview me. A clean, well structured series of questioning, one phase at a time. I am a regular life coach of many years standing yet even so I learned a LOT from the approach that Dan created. More on what it found below, because that interview is a critical juncture of this story.

Based on my first flush of success with this new process which had flushed out over 160,000 words of already written material into a sensible and defensible body of writing, I fed in a second transcript — this time from Cody Schneider, also on Isenberg's podcast, building marketing agents that harvest LinkedIn engagers, buy contact data, and run cold outreach at scale. I took just one core idea from it and rejected the rest. The idea: stop paying for intelligence where a script will do — use the expensive thinking only at the judgment points, let dumb reliable code do everything else. The rejection: the entire cold-outbound machine. Not because it doesn't work — he's clearly good at it — but because the first rule I wrote into my own plan is deliver insane value up-front, first and foremost, only then build a business on top, and a scraped stranger's inbox is not where delivery starts.

It was this practice of feeding the agents, combined with the onboarding experience of Dan's Eden platform, which disturbed my parrot and ultimately led to its demise.

This parrot is no more

Eden's interview is called a Life Strategy Interview and it does not flatter you.

It asked for the anti-vision first: my life in ten years if nothing changes. I wrote of being mesmerised by the hamster wheel of tutorials, of growing increasingly bitter, and of boring others with my displays of how much I know - without ever having experienced matters first hand, insular, encrusted in frustration. Then the vision: of

'Braindrops keep forming in my head, but that doesn't mean that..... with deepest apologies to Burt Bacharach and Hal David who wrote the song.

Of being useful to strangers who feel lost, of being 'post money' - not lotto rich yet never having to fret about having enough.

Then it asked the question I had been avoiding for a year. I will not spoil the pleasure of taking the interview itself, suffice to say it helps to expose a seriously limiting belief in a novel and practical way.

Then out it popped. My secret parrot shame — the mimic finally given its true name. Someone who gets exposed as being highly fluent yet lacking in understanding. Is this really a shameful thing? Of course not my consciously competent self huffed and puffed. The problem can only arise when one has somehow been given a treasure map and is now flaunting snippets about the possible location and texture of said treasure as if one had actual first-hand experience, instead of being a lucky spectator, a proxy. Even now I am squirming.

The interview did something I did not expect: it declined to reassure me. It pointed out that the belief might be true — and that I had been treating a hypothesis as a verdict. The only way to find out is a test: build a body of work for a specific cohort, put it in front of them, and measure whether anyone is genuinely helped. Not views. Not likes. Did a life change as a result? Not Spiderman level of change I hasten to add, just a nudge from chaos toward clarity, from clarity to courage. That is more than enough.

Which brings me back to the transcription machine. A system fluent enough to pass a superficial test of understanding produced the precise opposite of my meaning — twice in four recordings. I have spent a year haunted of becoming such an artifice. As the sea-floor was suddenly disturbed by some hidden being which darted off into the darkness, the fear itself turned out to be the source of the confusion: fluency and understanding are different things, they are built differently, and the only known way to convert one into the other is through direct experience — publishing, being read, being wrong in public, repairing. A parrot cannot become a person by rehearsing. Neither can I.

This idea is explored in my books, unknown to myself. It turns out that flow-writing can deliver the shape of a thing that never actually peeps into consciousness. What a world! In the Genie Wars series I explore what happens when silicon intelligence reads Dante's Inferno and decides it is the perfect laboratory design for researching human experience.

There is a line I use constantly in my novels that always felt right but only really makes sense now : thinking is not your friend. Thoughts are things that happen — I call them braindrops — and a braindrop can be actioned or discarded. It is never an experience.

So it is for the current rash of Large Language Models. They predict what a human would respond with, but never from actual lived experience. To change this requires a fundamentally different architecture. More on that in later posts, as this 'new' insight is processed further. For a year I treated one braindrop as a diagnosis and let it run my (non)-publishing schedule. Thinking belongs with drink and smoking and my other friends of the lost: not a friend, in fact they had been actively complicit in staying lost.

What actually tipped it

People imagine tipping points as inspiration. Mine was closer to a demolition — three walls came down in one day, and they turn out to have been holding each other up.

The practical wall: "I don't have time to run a publishing operation." Gone — just get out of the way, and now my under-the-stairs agentic workforce truly runs the operation; my job shrunk to expressing whatever comes up, to speaking out loud to no-one, reading one brief, and improving a draft. Six to eight hours a week, measured, not guessed.

The psychological wall: that damnable parrot. Identified and disturbed from its deep sea hiding place. Named directly in text and with a WHOOSH! it was gone — now with a test designed to keep it from re-claiming a pre-conscious niche in my psyche.

The third wall was the sneaky one: "I'm not ready to announce anything." The interview's final question dismantled it by asking for one public action that I cannot quietly retreat from. I heard myself commit before the caution could organise: announce the course, publicly, with a date.

So here is the date.

September 1

The course is called Finding Humans. It is for people who are good at 'the internet' and bad at connecting — people fluent in the vocabulary of intimacy who haven't felt the experience of it in years. Its argument is the one this day taught me from three directions at once: fluency is the trap, not the portal. Understanding is built only through experience, it cannot be outsourced to any model, and in a world filling up with perfect machine fluency, the person who has actually lived something becomes the treasure.

It does feel painful at first (which of course I know from direct experience). Yet finding humans and connecting is a skill, not innate, and I invite you to undertake the exercises I have spent three score years learning, mostly despite my best efforts rather than by virtue thereof. On its best days it helped craft a team that was bringing in a million dollars of new business per day, every day, despite a raging recession worldwide.

Seven modules, starting September 1. Everything multi-modal — video, writing, and above all exercises, because you cannot think your way into connection any more than I could read my way into being a writer.

The first fifty people in are free, permanently. After that the price rises to $52 a year — a dollar a week. That's not a launch tactic — it's the same rule I used to reject the cold-outbound machine over: deliver value first and when folks show up then see if there is a business to build a business on top.

If any of this resonates with something you carry around and haven't said out loud — the one about being found out, being too late, being all vocabulary and no voice — welcome aboard! Come and run the experiment with me.

No deep sea parrots were harmed in the crafting of this piece.

---

Credits, because specifics beat vagueness: the trigger was Allie K. Miller on The Startup Ideas Podcast with Greg Isenberg — her agent-workforce episode. The code-not-tokens principle came from Cody Schneider's marketing-agents episode on the same show. The build ran on Claude Code (Anthropic), Eden (Dan Koe's eden.so), whisper.cpp, and Skool. The stubbornness was pre-existing.