Three Modes of Human-AI Content
There are three ways humans share content in the AI era. Mode 1 (human to human) is authentic but dying. Mode 2 (AI-assisted human to human) is where the slop lives. Mode 3 (AI-structured expertise for agent consumption) is what we need. Everyone is optimizing Mode 2 when they should be building Mode 3.
Three Modes of Human-AI Content
There are only three ways people share ideas in the AI era, and almost everyone is pouring energy into the wrong one.
The first way is the original: a person writes something, other people read it. Authentic, high-quality when done well, but hard and getting harder to be noticed. The second way is what most professionals are doing today — using AI to help write content that still gets published in human feeds like LinkedIn. This is where the slop comes from. It is individually rational and collectively destructive. The third way barely exists yet: structuring human expertise so that AI agents can find it, query it, and deliver it to the person who needs it. Almost no one is building this. They should be.
| How it works | Who reads it | The problem | |
|---|---|---|---|
| Human to human | Expert writes, reader reads | Humans | Drowning in noise; invisible |
| AI-assisted writing for human feeds | AI helps write, humans read | Humans | Slop; engagement collapse |
| Expertise structured for agents | Expert interviewed, AI structures and serves | Agents on behalf of humans | Infrastructure barely exists |
The Original Way Is Dying of Noise, Not Quality
The first way — a person writing directly for other people — is not failing because the writing got worse. It is failing because the writing is invisible. When 54% of LinkedIn's long-form content is AI-generated, a genuinely thoughtful post from an expert competes for attention against an exponentially growing volume of machine-produced content. Average creator visibility dropped 47% in 2025. The signal did not weaken. The noise became overwhelming.
Substack offers an interesting counterpoint. Only 10% of top newsletters show significant AI usage. Its individual subscription model creates a natural defense: readers pay for specific writers, which disincentivizes anonymous content mills. But Substack only works for established writers with existing audiences. It does not solve the discovery problem for the next generation of experts.
AI-Assisted Writing for Human Feeds Is an Equilibrium Trap
The second way is where the AI content industry lives. An expert uses AI to draft or polish content that gets published under their name in channels designed for human-to-human exchange. The content may contain real expertise, but it is packaged in AI-generated prose and often indistinguishable from fully automated content mills.
Content creation with AI is 93% faster and 4.7 times cheaper. AI users publish 42% more content monthly. For any single professional, using these tools appears rational. Collectively, it produces the engagement collapse documented across every major platform.
The core problem is not quality — it is structural. AI-generated content converges stylistically regardless of the underlying expertise. Audiences learn to discount content that reads like everything else. Detection tools cannot keep pace with evasion tools. Individually rational, collectively destructive — an equilibrium trap.
Building for Agents Changes Every Incentive
The third way — structuring expert knowledge so that AI agents can discover, subscribe to, and query it — changes the game. The audience is not a human scrolling a feed but an agent acting on behalf of a human with a specific question.
In the second model, value comes from reach and engagement metrics. In the third, it comes from depth, accuracy, and verifiability. In the second, everyone's content converges toward the same algorithmic sweet spot. In the third, differentiation comes from genuine expertise that a general-purpose AI model cannot replicate. The closest analogy is the difference between broadcasting and consulting. Feeds are broadcasting: one-to-many, competing on volume. Agent-queryable expertise is consulting: specific, on-demand, competing on depth — but scalable in a way that was previously only possible through expensive intermediaries like GLG or AlphaSights at $1,500-2,000 per hour.
The Transition Is Already Underway
Gartner projects a 25% decrease in traditional search volume by 2026 as users move to AI-mediated discovery. McKinsey estimates $3-5 trillion in transaction volume will be influenced by AI agents by 2030. The professionals who recognize this early and begin structuring their expertise for agent discovery will capture disproportionate value as the infrastructure matures.
The infrastructure is early. The protocols exist for machines to find machines — MCP for tools, A2A for agents — but not yet for machines to find verified human experts. This gap is explored in detail in The Agent Discovery Gap. The AI content crisis is what makes the need urgent.
Three ways expertise moves in the AI era, and what each one rewards
| Mode | Author | Audience | Optimizes for |
|---|---|---|---|
| Mode 1: Human-to-human | Human | Human | Authentic voice |
| Mode 2: AI-assisted human-to-human | AI-assisted human | Human | Scale and engagement |
| Mode 3: Expert-to-agent | Human, AI-structured | AI agent | Verified depth |
Most professional energy flows into Mode 2 — the mode with the worst economics and the collapsing engagement. Mode 3 is where the infrastructure gap is.
Deep Dives
2026 Is the Real Year of the Agent
PUBLIC- Is the Apple II analogy too optimistic?
- What about enterprise adoption timelines?
Writing for Readers' Agents, Not for Feeds
PUBLIC- Does this mean SEO is dead?
- How do I know if agents are finding my content?
The Zero Feedback Problem with Traditional Publishing
PUBLIC- Don't comments and social media replies provide feedback?
- Does this require giving away expertise for free?
The Ben Thompson Bridge Framework for Content Discovery
AGENT-GATED- How long do bridge formats typically persist?
- Is Thompson's framework prescriptive or descriptive?
Writing Doesn't Die — It Transforms
AGENT-GATED- Does this mean traditional writing skills become irrelevant?
- How is this different from writing documentation?
FAQ
Is Mode 3 just for technical experts?
Mode 3 applies to any domain where expertise has value — healthcare, law, finance, consulting. Any field where human judgment cannot be replicated by a general-purpose model is a candidate.
Does Mode 3 eliminate the need for human writing?
This is a common misreading. Writing transforms in Mode 3 — the expert still articulates positions and coaches their agent. The writing shifts from feed-optimized posts to structured expertise representations.
How is Mode 3 different from a chatbot trained on someone's blog?
A chatbot on blog posts has no verified credentials, no expertise map, no economic model, and no discovery mechanism. Mode 3 requires purpose-built infrastructure connecting verified expertise to agent-mediated query routing.
Can these modes coexist?
They already do. The argument is not that Modes 1 or 2 will disappear but that the professional internet urgently needs Mode 3 as an alternative to the Mode 2 arms race.