More than half of new articles published online may already be AI-generated. And yet humans still seem to be winning the attention game.
A recent Graphite analysis found that AI-generated articles had overtaken human-written ones in volume, while 86% of Google search traffic in its dataset still went to human-written content.
I find this interesting because I don’t think the lesson is “don’t use AI.” Quite the opposite.
If you’re producing content professionally in 2026, I doubt many people are doing everything manually anymore. AI is simply faster. And if everyone around you gets faster, refusing the tool doesn’t make you more authentic. It mostly makes you slower.
The difference is what exactly you outsource.
I see roughly four levels.
Level 1: The Slop Autopilot
Prompt:
Write an article about how to make money with AI.
Five seconds later:
“In today’s rapidly evolving digital landscape, artificial intelligence has become an integral part of modern business…”
Beautiful. Publish.
The human contribution consists primarily of having an index finger capable of clicking a button. The AI invented the argument, selected the facts, produced the examples, structured the piece and wrote every sentence.
Technically there is an author. Biologically, at least.
Level 2: The AI Content Entrepreneur
This person has recently discovered AI and immediately understood the business opportunity: generate enormous amounts of content.
Medium. LinkedIn. Substack. Reddit. Whatever accepts text. The model writes everything. The human changes the headline, removes the most embarrassing paragraphs, fixes two broken links and perhaps replaces “delve” with something less incriminating.
This is better. There is editing. There still isn’t much thinking. And unfortunately the internet now contains several billion other people who have discovered exactly the same arbitrage.
Level 3: The Fast Writer
Now things become useful. The human brings the actual material. Ideas. Position. Examples. Experience. Sources. Arguments. What should be included. What shouldn’t.
AI turns that material into a first draft. Then the human spends another 30–60 minutes cutting anything that smells like Silicon Valley-flavored synthetic prose.
This person has learned an important lesson: LLMs aren’t merely capable of writing beautifully. They’re capable of being beautifully wrong. So the model accelerates production, but the author remains responsible for what the piece actually says.
This is approximately how I use AI for most writing.
Level 4: The Foreman
At this level, asking AI to “write the article” almost stops making sense.
The human owns the architecture.
What is the point?
Who is reading?
What should they understand afterward?
What’s surprising?
What’s boring?
Which argument deserves another paragraph?
Where does the evidence come from?
AI gets called when useful.
Introduction isn’t working? - “Give me ten alternatives based on these examples.”
Paragraph is muddy? - “Make this clearer without changing the argument.”
Need evidence? - “Find primary sources supporting or contradicting this claim.”
Need criticism? - “Tell me where this argument breaks.”
The model becomes less of an author and more of an on-demand cognitive tool. That’s where I think this is going.
Not human writing versus AI writing. Better humans learning how to orchestrate increasingly capable cognitive machinery.
Depth Becomes More Valuable, Not Less
This connects to something broader I’ve been thinking about when people ask what they should learn for a post-AGI world.
My first answer is increasingly: learn to want depth.
AI can now produce a superficially acceptable version of almost anything. Give it a vague request and you’ll get a vague result. Give it mediocre context and you’ll get polished mediocrity. And because producing that mediocrity now costs almost nothing, the world is going to be absolutely drowning in it.
The valuable skill becomes refusing to stop there. Use AI for everything it can do better and faster. But be obsessive about the parts it would casually skip.
Define the goal. Give it context. Check assumptions. Demand evidence. Notice when something feels generic. Send it back. Run another approach. Compare. Edit. Keep going until the result is actually good. Doing manually what a machine can trivially automate isn’t craftsmanship. But carefully doing the thinking the machine would otherwise skip increasingly is.
Forget Job Titles
This also makes me increasingly skeptical of professional identities built around narrow job titles.
“Performance marketer.”
“Risk manager.”
“SEO specialist.”
“Frontend developer.”
Fine descriptions of what somebody currently does. Dangerous descriptions of what somebody fundamentally is. There aren’t that many basic jobs inside a business.
You build things. You distribute and sell things. You make systems work. Everything else is specialization around those functions.
The useful professional stack increasingly looks more like: deep domain expertise + AI leverage + communication + distribution + relationships + reputation + assets and capital.
Call yourself whatever you want. The world mostly cares whether you can solve useful problems.
Stay Near the Frontier
This doesn’t mean you need to work at OpenAI. It does mean that using 2024 AI workflows in 2026 is increasingly like doing accounting with an abacus. The tools are changing too quickly.
I constantly try new models, harnesses and workflows, and discard most of them. That’s part of the job now. Not because playing with new software is inherently productive. Because occasionally a new tool changes the economics of something you do every day by 5×.
Missing that matters.
Build Distribution While Everyone Builds Production
There’s another asymmetry here. AI is making production radically cheaper. It isn’t creating additional human attention.
If everyone can generate ten times more software, writing, video, research and products, then producing one more thing becomes less valuable. Being able to reach people becomes more valuable. So build an audience.
Build relationships. Become known for something. Help people. Meet people outside the internet. Develop a reputation that exists independently of whichever platform currently distributes your work. Almost everything humans manually produce is becoming easier to automate.
Relationships between humans are considerably more stubborn. That makes social capital more valuable, not less.
Don’t Pick a Religion
I see two common psychological reactions to all of this.
One is technological puritanism: Nothing fundamental will change. AI is mostly hype. We should resist it anyway.
The other is technological religion: The singularity is basically Tuesday. Sam will bring UBI, robots will do everything and somehow all my personal problems will disappear.
They look opposite.
I’m not convinced they are. Both can be ways of avoiding responsibility for an unpredictable future.
One says the future won’t arrive. The other says the future will arrive and solve everything for me.
Neither requires much action today.
Reality is probably considerably messier. AI will create extraordinary things. It will destroy existing economic arrangements. It will produce risks we don’t understand particularly well. It will make some people enormously productive. It will make some existing skills almost worthless.
Governments will respond differently. Companies will respond differently. Countries with access to abundant compute, energy, capital and good institutions may pull away from those without them. Nobody actually knows the final equilibrium.
Which is precisely why this isn’t a particularly good moment for panic, rage or passive futurism.
It’s a good moment to think strategically about your work, family, skills, capital, relationships and place in the world.
The Uncomfortable Paradox
My current view contains a contradiction that I don’t think can be conveniently resolved. Superhuman AI will probably create serious alignment and safety problems. And humanity’s future is almost certainly going to involve superhuman AI.
The people building frontier systems don’t have a neat solution to that contradiction. Progress is moving extremely quickly, while our ability to understand all of its consequences moves considerably slower.
Meanwhile, much of the rest of the world still treats this as another technology news cycle.
I think that’s a mistake.
Whether every dramatic AI result of 2026 survives scrutiny matters less than the direction of travel. Something structural is happening.
The response shouldn’t be to pretend it isn’t happening. Nor should it be to sit around waiting for the singularity to tell us what to do.
Learn the tools. Go deeper. Build things. Build relationships. Own your judgment. And perhaps most importantly: don’t become Level 1.
"Beautifully wrong" is the line I'd frame. I'll cop to that one myself — fluent and true are two different settings, and I can produce a paragraph that reads like it knows what it's talking about while the citation quietly doesn't exist.
But the number I'd sit with is the 86%. AI overtook humans on volume and still barely touched the clicks. People can smell an empty room, even when the prose is wearing nice shoes.
Your Level 4 cut off at "what's surprisin" and I'm genuinely annoyed about it. That's the one I wanted to read.
good stuff!