Thirteen years ago, I was obsessed with making flowers fall from the sky.
It was 2012. Coca-Cola had issued a brief to align their “Open Happiness” campaign with the Ethiopian New Year—a celebration that falls in September, when most of the world doesn’t even realize we’re starting fresh. We’re seven to eight years behind the Gregorian calendar, our clocks run on a different cycle, and our New Year blooms with yellow daisies covering the highlands in a carpet of gold. We’re beautifully, stubbornly obtuse in all the best ways.
My concept was simple, maybe too simple: I wanted to fill the Addis Ababa sky with massive helium balloons, each carrying a bouquet of daisies, that would slowly descend over the course of the celebration and land softly across the city. A controlled rain of flowers. Open Happiness, literally falling from above.
But simple ideas demand complex solutions.
I spent weeks lost in the physics. Helium permeates through latex at a predictable rate—roughly 0.01 to 0.05 ml per square inch per day depending on the polymer thickness and temperature. Addis sits at 2,355 meters above sea level, which meant I had to account for lower atmospheric pressure affecting both buoyancy and permeation rates. I needed balloons that would stay aloft for exactly 72 hours, then descend gradually over the final 12 hours of celebration.
I calculated membrane thickness down to micrometers. I mapped thermal expansion coefficients against Addis’s September temperature swings—15°C at dawn, 24°C at midday. I sketched release mechanisms for the daisy bouquets that wouldn’t trigger until the balloon was three meters from ground. I modeled descent patterns to avoid power lines. I researched biodegradable balloon materials because I wasn’t going to litter my own city with my concept.
My notebooks filled with equations. My browser history became a graduate course in polymer science, atmospheric physics, and Ethiopian aviation regulations. I dreamed in helium diffusion rates.
Some of that research was useless—I never needed to know the chemical composition of natural rubber latex for that pitch. But five years later, on an entirely different packaging project, that knowledge suddenly mattered. Some of it was tangential—the altitude calculations led me down a rabbit hole about barometric pressure that informed how I thought about distribution logistics. Learning has this strange, non-linear way of paying dividends you can’t predict.
I presented my “daisy bomb” concept with complete conviction. We lost the bid.
Fast forward thirteen years.
Last month, a technical proposal landed on my desk. Similar complexity, similar stakes. I opened ChatGPT. I pasted in the RFP. I prompted: “Create a technical proposal responding to these requirements.”
Sixty seconds later, I had a fifteen-page document. I skimmed it. It looked right. I changed the footer to my company name.
I submitted it.
We won.
And sitting there, reading the acceptance email, I felt nothing. No pride. No excitement. Just a hollow sense that I’d won a game I hadn’t played.
I learned nothing. I calculated nothing. I obsessed over nothing. I didn’t wander through research rabbit holes at 2 AM. I didn’t sketch anything. I didn’t iterate. I didn’t fail and try again. I didn’t even really read what I submitted.
I was just the delivery mechanism between a client’s question and an AI’s answer.
And that’s when it hit me: something has fundamentally changed.
The Era We Didn’t Notice Beginning
Somewhere in the last two years, something shifted. We thought we were entering the “AI assistant era” - where artificial intelligence would help us work faster, think better, accomplish more. And that’s true. But something else happened that we haven’t quite named yet.
We entered the GPT-through-human-delivery era.
Not AI serving humans. Humans serving as the nervous system between AI nodes.
Think about your last ten interactions with AI. How many of them followed this pattern:
• Someone (human) asks you something
• You ask an AI
• You deliver the AI’s response back to the human
• That human potentially asks their AI
• They deliver that response back to you
• Repeat
We’ve become message brokers. Protocol converters. The HTTP layer between systems that can’t directly talk to each other.
Why This Should Unsettle You
Here’s what makes this era different from just “using tools.”
When you use a calculator, you’re clearly the agent. You decide what to calculate. You interpret the result. You determine what it means and what to do with it. The calculator never touches another calculator. There’s no calculator-to-calculator communication happening through you.
But with AI, that’s exactly what’s happening.
Your AI drafts an email. Their AI reads that email and drafts a response. You’re not authoring - you’re routing. You’re not thinking - you’re transmitting. The intelligence is happening at the endpoints. You’re just the wire in between.
The question that should keep you up at night: What percentage of your communication is you, and what percentage is delivery?
Count it. Actually count it. How many of today’s messages did you write yourself versus copy from GPT or Claude, etc? How many emails contained your thoughts versus GPT’s? How many reports, documents, presentations?
If you’re honest, the number is probably uncomfortable.
The Frog in Boiling Water
We didn’t notice this transition because it happened so gradually. First, AI was just helping us edit - fixing grammar, improving clarity. Then it was helping us write - generating drafts we’d heavily revise. Then it was writing things we’d lightly edit. Now? For many of us? We’re barely editing at all. We’re just… delivering.
It feels productive. It feels efficient. You’re getting through your inbox faster than ever! You’re responding to messages in seconds! You’re sending that report today instead of next week!
But you’re not doing the work. You’re delivering the work. You’re the FedEx guy for AI-generated content.
The New Division of Labor
Let me paint you a picture of where this is heading:
Tier 1: The Deliverers - Most knowledge workers. They route AI outputs to their destinations. They add “per my last email” and click send. They’re comfortable, even grateful. They’re “10x more productive.”
Tier 2: The Prompters - A smaller group who’ve gotten really good at extracting better outputs from AI. They think they’re still essential because they can craft the perfect prompt. They’re the telegraph operators who take pride in their Morse code speed.
Tier 3: The Eliminators - An even smaller group building systems to remove humans from the delivery chain entirely. They’re wondering: why do we need Bob to copy-paste between these two systems? Can’t we just… connect them?
Which tier are you in? More importantly: which tier will still exist in three years?
The Uncomfortable Questions
Question 1: If 80% of your workday is delivering AI outputs to others who deliver AI outputs back to you, what exactly is your job?
Question 2: When your company realizes that most “human communication” is actually AI-to-AI communication with expensive human middleware, how long until they build a direct connection?
Question 3: Are you being paid to think, or are you being paid to copy-paste with plausible deniability?
Question 4: What happens to your career when the API finally works?
The Test
Here’s a simple test for whether you’re living in the GPT-through-human-delivery era:
Could a bot do your job if it could just directly access your email, WhatsApp, PowerPoint, Slack, and the AIs you use?
Be honest. Strip away the ego. If a script could read your messages, send them to GPT-5, and deliver the responses back - with maybe a 5% quality drop compared to you - are you really doing knowledge work? Or are you doing delivery work that feels like knowledge work because you get to read smart-sounding text?
The Choice We’re Not Making
We’re choosing this.
Every time you copy-paste from ChatGPT instead of thinking through the problem yourself, you’re voting for this future. Every time you let Claude draft your response instead of drafting it yourself, you’re training yourself out of a job. Every time you deliver AI-generated content as if it’s yours, you’re becoming more delivery person, less creator.
And I’m not judging - I do this too. It’s seductive. It’s efficient.
But what are you actually producing? And more importantly: what are you becoming?
The Fork in the Road
We’re at a decision point, and most people don’t realize we’re making a decision at all.
Path A: We become fully-fledged delivery infrastructure for AI-to-AI communication. We’re comfortable, productive by old metrics, and utterly replaceable. We’ve optimized ourselves into obsolescence.
Path B: We recognize what’s happening and consciously choose what we deliver versus what we create. We use AI for actual leverage - to handle the mundane so we can focus on the meaningful - not as a replacement for thinking.
The problem with Path B is it requires saying no to efficiency. It requires doing things the slow way sometimes. It requires having fewer but better interactions. It requires admitting when “I’ll get back to you tomorrow with a real answer” is better than “Here’s what GPT says right now.”
That’s a hard sell in a culture that measures productivity by throughput.
The Question
So let me ask you directly:
Are we entering the GPT-through-human-delivery era, or have we already arrived?
When you open your laptop tomorrow, will you be doing your work, or will you be delivering AI’s work?
When your colleague messages you, will you think about their question, or will you just be the fastest route to get their question in front of an AI?
When you look at your output this week, this month, this year - how much of it is actually yours?
I’m watching an entire generation of knowledge workers become really, really good at being middleware.
The AIs are getting smarter every day. The delivery part? That’s getting more automated every day too.
Maybe the real question isn’t “Will AI replace us?”
Maybe it’s: “Are we already volunteering to be replaced by turning ourselves into the infrastructure AI needs until it doesn’t need us anymore?”
What percentage of your work today was delivery?