Skip to content
All writing

Pay Tokens for Attention

Human taste decides the direction. AI supplies the surface area. Engineering makes it compound.

Hongkai He 15 min read
  • #ai
  • #growth
  • #distribution
  • #causally

Growth is not the marketing department’s side task. For an AI-native product studio, growth is part of product design, engineering design, content design, and organizational design. If you build products in an AI-native way but grow them in a pre-AI way, you leave most of your advantage on the table.

What follows isn’t a rigid playbook. It’s a mindset and an operating framework. The core idea:

Pay tokens for attention — spend compute to turn data and context into useful, personalized, distributed artifacts that earn discovery, clicks, shares, replies, and conversion.

But before the AI-native part, we have to respect the fundamentals.

I’ll use one running example throughout: a menu-translation app for travelers — photograph a foreign-language menu, get back translated, visualized dishes. It’s concrete enough to make the abstractions land.

The fundamentals don’t change

A product doesn’t grow because you “do marketing.” It grows because you create a loop:

Product → user value → user action → distribution → more users → feedback → better product.

Every launch and growth effort should be read through a practical founder/operator map:

LayerCore questionWhat good looks like
PositioningWho is this for, and why now?A specific user in a specific high-pain moment
ActivationHow fast does the user feel the “aha”?The value is experienced within seconds or minutes
DistributionWhere do users come from?Clear channels, not vague “awareness”
RetentionWhy would users come back?The product solves a recurring need, not one-time curiosity
Referral / loopWhy would one user bring another?Sharing is part of the user’s natural workflow
MonetizationCan growth be funded sustainably?CAC, conversion, retention, and margin can work together

For the menu app, the basic growth logic shouldn’t start from “we need marketing.” It should start from:

QuestionMenu-app answer
Who is this for?Travelers, international students, food explorers — anyone facing a text-only or foreign-language menu
What is the painful moment?Sitting in a restaurant, seeing unfamiliar dish names, not wanting to order blind
What is the aha?Photograph a menu → see translated, visualized dishes
What creates sharing?A visualized menu is naturally useful to your dining companions
What creates retention?Saved menus, food memory, trip use, repeated restaurant discovery
What creates monetization?Subscription, credits, travel pack, premium visualizations, B2B restaurant/menu partnerships

Traditionally, growth channels are grouped into organic and paid:

Growth typeWhat you pay withExamplesBest use
OrganicHuman time, creativity, product work, community trustSEO, content, referrals, PR, Product Hunt, Reddit, social, product-led loopsDiscover durable growth loops and earn trust
PaidMoneySearch ads, paid social, influencer sponsorships, affiliate, retargetingBuy speed, test messages, scale proven loops

But the more useful distinction is often demand capture vs. demand creation:

ModeUser stateExample channelMenu-app example
Demand captureAlready knows the problem and is searchingGoogle, App Store, SEO, search ads”menu translator app,” “Japanese menu translation,” “how to read a French menu”
Demand creationFeels the pain but hasn’t named the categoryTikTok, Reels, YouTube, PR, creator content”This menu has zero pictures. Let’s use AI to see what I almost ordered.”

These fundamentals are still true in the AI-native era. AI doesn’t remove the need for positioning, activation, retention, referral, or unit economics. AI changes the production function behind them.

The mindset shift: pay tokens for attention

Traditionally, organic growth is “free” only in the sense that you don’t directly pay the platform. In reality, you pay with human labor.

AI changes this equation:

Growth production modelInputOutput
Traditional organicHuman labor + channel knowledge + creative workAttention
PaidCash + targeting + creativeAttention
AI-nativeData + context + models + workflow automation + human judgmentAttention

This is what I mean by pay tokens for attention. Tokens don’t buy attention directly the way ads do. Tokens buy the ability to create, adapt, personalize, monitor, respond, and test at a scale human labor could never support.

Old constraintAI-native replacement
Human writer timeToken-generated pages, answers, captions, scripts, guides
Human designer timeAI-generated visuals, layouts, thumbnails, cards
Human researcher timeAgents monitoring keywords, communities, trends, competitors
Human social manager timeAI-assisted contextual replies, drafts, routing, follow-ups
Human SEO laborProgrammatic page generation + AEO/GEO optimization
Human ad creative teamCreative variant factory
Manual analyticsAutomated experiment review and next-test suggestions

This is not merely “AI helps us write copy faster.” That’s the shallow version. The deeper version is:

Compute becomes a new marginal cost of growth.

You can spend tokens to turn raw data and context into useful artifacts that travel through the internet.

From programmatic SEO to AI-native distribution

Programmatic SEO was an early version of this mindset.

EraWhat scaledExample
Manual contentHuman-written pages100 blog posts
Programmatic SEOTemplate + structured databaseYelp/Zillow/TripAdvisor-style long-tail pages
AI-native distributionData + reasoning + multimodal generation + personalization + feedbackVisual menu pages, dish explainers, restaurant guides, contextual answers, short videos

AI expands “template + database” into something much richer:

Traditional programmatic pageAI-native artifact
Static text pageMultimodal page with images, explanations, translation, FAQs, schema
Same template for every entityContext-aware output by restaurant, dish, cuisine, city, user intent
SEO-onlySearch + social + community + image discovery + AI-agent readability
One-way publishingFeedback loop from clicks, uploads, shares, saves, payments

For the menu app, this could look like:

Raw inputAI transformationDistributed artifactGrowth value
Restaurant/menu dataOCR, dish understanding, translation, visualizationVisual menu pageSEO, sharing, conversion
Dish namesIngredient/cuisine explanation + generated imageDish explainer cardSocial, image search, education
City/area restaurant listCluster by cuisine, traveler intent, dietary needsCity food guideLong-tail search and travel discovery
Reddit food/travel questionContextual answer + optional visualized exampleHelpful reply draftCommunity discovery
Uploaded user menuClean share pageA “see what we’re ordering” linkReferral loop
Food trendHook / script / storyboard generationTikTok/Reels/ShortsDemand creation

The loop becomes:

Menu/restaurant data → AI-visualized artifact → search/social/community distribution → user upload/share/save → feedback data → better artifacts and targeting.

This is the core of AI-native distribution.

The token-to-attention stack

Think about AI-native growth as a stack.

LayerQuestionExample
1. Data substrateWhat raw material do we have or can legally access?Restaurant names, menus, dish names, user uploads, public food questions, city/cuisine data
2. Artifact factoryWhat useful outputs can AI produce?Visual menus, dish cards, ordering guides, short video scripts, community answers
3. Distribution agentsWhere should each artifact go?SEO, TikTok, Reddit, Product Hunt, travel blogs, AI-answer-readable docs
4. Feedback engineWhich artifacts actually work?Impressions, clicks, uploads, shares, saves, signups, payments, repeat use
5. GuardrailsWhat should we not automate or publish?Spam, fake experience, copyright issues, hallucinated claims, platform TOS violations

The practical flow pairs a system action with a human responsibility at every step:

StepSystem actionHuman responsibility
1Monitor opportunitiesDefine what counts as a real opportunity
2Generate artifact candidatesDefine creative direction and quality bar
3Score quality and riskSet thresholds and review edge cases
4Distribute or queue for reviewProtect platform trust and brand tone
5Measure downstream behaviorInterpret what the numbers actually mean
6Generate next experimentsDecide what deserves more scale

The goal is not “infinite content.” The goal is infinite specific usefulness under quality control.

Concrete AI-native growth patterns

Here are the patterns worth using as inspiration.

PatternWhat it meansExampleMain risk
Programmatic utility pagesGenerate useful long-tail pages from structured entitiesOne visual menu page per restaurant / menu / dish / cityThin SEO spam if pages lack real utility
Contextual community participationFind real questions and answer with useful contextReply to “what should I order?” posts with visual guidanceSpam, undisclosed promotion, platform bans
AI creative factoryGenerate many video/post variants and test hooks100 hooks around ordering blind, travel menus, dish surprisesLow-quality AI slop, weak brand taste
Personalized landing pagesAdapt pages by user intent / audienceVegan traveler, Japan trip, date-night ordering, student abroadOver-personalization or hallucination
AEO/GEO contentMake content readable by AI answer engines and agentsStructured markdown, FAQs, schema, clear claims, source contextGeneric content that agents ignore
Shareable artifact loopProduct output itself spreadsA visualized menu link shared with friends before orderingOutput not beautiful / useful enough to share
Autonomous experiment generationAgents create and test many small experimentsHook × audience × format × CTA variantsOptimizing vanity metrics instead of retained users

A good AI-native growth system should always ask:

  • Is this artifact genuinely useful? Utility protects you from spam dynamics.
  • Is it specific to a real context? Specificity beats generic AI content.
  • Is it connected to product activation? Attention without activation is noise.
  • Can users share or reuse it? Sharing creates compounding loops.
  • Can we measure downstream value? You need learning, not just output volume.
  • Can this be safely automated? Trust is worth more than volume.

AI-native growth is not humanless growth

This thesis is not “AI replaces the growth team.” It’s the opposite. AI-native growth turns humans from task executors into opportunity spotters, taste setters, workflow architects, and judgment owners.

The new division of labor:

Human instinct identifies raw opportunity → human taste defines creative direction → AI generates high-volume artifacts → human/AI quality systems filter outputs → AI-native engineering automates the workflow → agents distribute and test → humans interpret feedback and update strategy.

Or shorter:

Instinct → Taste → Process → Automation → Distribution → Feedback → Judgment.

The human role moves upstream and becomes higher-leverage.

Human roleCore jobWhy it matters more in AI-native growth
Opportunity spotterNotice underserved, high-pain moments before they’re obviousAI can scan, but humans recognize meaning, urgency, and market asymmetry
Creative framerTurn pain into a compelling hook, story, or memeAI can produce options; humans sense what feels fresh, native, emotionally true
Taste setterDecide what looks good, trustworthy, premium, or cringeLow-cost generation makes taste the bottleneck
AIGC workflow designerBuild the right model / tool / prompt / control pipelineOne-off AI output is easy; repeatable quality is hard
AI-native engineerAutomate the workflow into a reliable systemGrowth experiments only compound when they become infrastructure
Channel-native operatorUnderstand platform culture and trust thresholdsReddit, TikTok, SEO, and Product Hunt each have different norms
Judgment ownerDecide what not to publish, fake, claim, scrape, or automateLow marginal cost creates high temptation and high risk
Feedback interpreterRead weak signals and update strategyMetrics show what happened; humans infer why

The key sentence:

AI scales execution; humans decide what is worth scaling.

What it means for each function

This isn’t only for “growth people.” Whatever your function, the question changes.

FunctionTraditional questionAI-native growth question
Founder / product ownerDoes this solve a problem?Does this product also create a compute-native growth surface?
EngineerHow do we build the product?What repeatable growth workflow can this product power?
DesignerIs the product / interface beautiful?What artifact from this product is worth sharing, indexing, remixing, or posting?
Growth / marketingWhich channel should we use?What data / context can we transform into useful distribution assets?
Content / brandWhat should we say?What creative grammar can AI scale without becoming generic or cringe?
OperationsHow do we keep work organized?How do we create review queues, QA rules, publishing controls, and feedback loops?

The highest-leverage person on a team is not necessarily the one who manually does the most tasks. It’s the one who can turn an insight into a repeatable, AI-powered loop.

Product selection should include growth surface

For every new product, evaluate not only feasibility and user pain, but also AI-native growth potential.

Evaluation questionWhy it matters
Does usage create shareable artifacts?Shareable outputs create referral and social loops
Does the domain have long-tail entities?Long-tail entities create SEO/AEO surfaces
Are there public demand moments?Public questions create contextual distribution opportunities
Can AI generate useful outputs at scale?Compute can replace large amounts of manual growth labor
Can outputs be personalized by context?Personalization increases relevance and conversion
Can we measure artifact-level performance?Measurement enables reinforcement learning
Does the product improve with more usage?Feedback loops create compounding advantage
Can we avoid spam / legal / trust blowback?A distribution advantage must not destroy brand trust

A product with a strong AI-native growth surface is not only easier to build. It’s easier to distribute.

Operating doctrine

Be aggressive in using AI for growth, but disciplined in how you do it.

PrincipleMeaning
Utility firstEvery generated artifact should help the user, not just promote you
Specificity beats volumeA specific answer / page / video for a real context beats generic content at scale
Human taste sets the barAI can generate, but humans define what is good enough
Automate after signalDon’t prematurely automate workflows before you know they work
Measure downstream valueOptimize for activation, retention, referral, and revenue — not vanity impressions
Respect platform cultureWhat’s welcome on one platform may be spam on another
Protect trustNo deceptive fake experiences, undisclosed promotion, or unsupported claims
Turn winners into infrastructureA successful experiment should become a repeatable system

The wrong interpretation: “AI lets us publish infinitely.”

The right interpretation:

AI lets us test and serve infinitely many specific contexts — but only publish where we create net utility.

Or even simpler: we don’t use AI to shout louder. We use AI to appear in more relevant moments with more useful artifacts than a human team could produce.

The final thesis

AI-native growth is not about replacing humans with bots. It’s about changing the growth production function.

InputWhat it contributes
Human instinctFinds raw opportunities before they’re obvious
Human tasteDefines what feels valuable, native, beautiful, and trustworthy
Human judgmentDecides what to scale and what to stop
Data / contextProvides raw material for personalized artifacts
Models / tokensGenerate and adapt content, visuals, answers, and experiments
EngineeringTurns one-off experiments into repeatable systems
Feedback loopsTeach us what actually earns attention and conversion

The winning team won’t be the one that generates the most content. It will be the one that best combines human insight with machine-scale execution.

This should become part of a studio’s product DNA: every product built with a compute-native growth surface, every growth experiment designed as a feedback loop, every person treating AI not as a shortcut for tasks but as leverage for opportunity, taste, workflow, and scale.

In one sentence:

Human taste decides the direction. AI supplies the surface area. Engineering makes it compound.