Video summary

RANKING How People Are Making Money With AI

Main summary

Key takeaways

Business

Overall premise

The video ranks AI-related business models by cashflow now vs. longevity/defensibility, critiquing “wrapper” and “content farm” approaches as vulnerable to platform competition (“Sherlocking”).

1) AI Wrappers (“digital landlord” model) — C+ tier

Core idea: Rent foundational AI capacity (e.g., via OpenAI API), wrap it in a nicer UI/workflow, and charge a premium.

  • Typical flow: user input → calls ChatGPT/LLM → displays output
  • Examples:
    • CalAI (calorie tracking): photo → LLM guesses calories (called “least accurate”)
    • PhotoAI (Pieter Levels): selfie upload → Stable Diffusion → professional LinkedIn headshots (claimed ~$138,000 at one point)
    • Formula Bot (David Bressler): text → LLM → Excel formula (claimed ~$226,000 MRR)

Main risks / why it’s not higher:

  • Sherlocked / platform replication: platforms add the feature natively, collapsing the wrapper’s value.
    • Example: pdf.ai “printing money” (PDF upload + Q&A) until ChatGPT added native PDF reading, “overnight.”

Ranking logic: good near-term monetization (“wave before it crashes”) but weak defensibility. C+.

2) AI Automation Agencies (AAA) — B tier

Core idea: Implement automation/workflows for businesses using tools like Zapier/Make, plus integration and maintenance.

  • Framing: “like an AI tradie” — connect existing tools, don’t build custom software.
  • Revenue model: monthly service fees; value tied to workflow complexity.
  • Example workflow: real-estate lead hits website → AI drafts personalized email → sends it → alerts sales on Slack.

Economics & typical costs mentioned:

  • Zapier/Make subscriptions: $20–$50/month
  • Client charges: hundreds to a couple thousand/month (depending on complexity)

Constraints / risks:

  • Not passive; you’re the bottleneck (APIs break → you fix; client deadlines).
  • Sherlocking risk: incumbents (e.g., Microsoft/Salesforce) embed similar automation, letting clients “press a couple of buttons” instead of paying an agency.
  • Market noise/toxicity: many “agency courses” are sold by people who haven’t earned from real delivery.

Ranking logic: easier to start than enterprise, but limited by time-for-money and defensibility. B.

3) Programmatic SEO (“infinite content glitch”) — D tier

Core idea: Extract competitor sitemaps + use AI generation to publish huge volumes of near-duplicate content, monetizing ads.

Process:

  1. Download competitor sitemap
  2. Generate “unique versions” for thousands of topics
  3. Publish at scale

Example case:

  • Jake Ward (“SEO heist”): exported competitor URLs → generated 1,800 articles in a few hours
  • Claimed results: 0 → 3.6 million organic visitors
  • Claimed ad arbitrage example: ~$1,000 spend → $50,000 ad revenue

Failure mode / why it’s D:

  • Google updates targeted at content farms:
    • Massive core update → content devalued
    • Site de-indexed; traffic “fell off a cliff” / “zero overnight”

Ranking logic: short-term arbitrage, long-term sustainability fails due to search enforcement and “cat and mouse” with Google. D.

4) AI Influencers (“simp economy”) — D tier

Core idea: Create AI-generated personas and monetize via subscriptions (Fanvue/OnlyFans-style).

Tech stack referenced:

  • Images: Midjourney / Stable Diffusion
  • Voice: ElevenLabs

Monetization:

  • Funnel followers/comments/DMs → paid subscription (e.g., Fanvue)

Examples:

  • Aitana Lopez: “almost 400K followers,” framed as AI persona; “thousands/month in brand deals”
  • Karen AI: trained on voice; “charged $1/minute”; first weeks: $71,000

Why it’s ranked low:

  • Platform risk: Meta shadowbans AI accounts that don’t self-label.
  • Commoditization: low barrier to entry—anyone can generate personas with tools.
  • Ethical risk highlighted: monetizing loneliness and automated emotional engagement.

Ranking: potentially high margins but high platform and competitive risk. D.

5) Enterprise AI Consulting (private “sovereign AI”) — S tier

Core idea: Build private AI systems for large companies that can’t use public LLMs due to privacy/compliance constraints.

Typical approach described:

  • Start with an open model (e.g., Llama 3)
  • Apply fine-tuning and RAG (retrieval augmented generation) to ground answers in private documents

Why it matters:

  • “In-house ChatGPT” experience without leaking secrets
  • Full context on internal work

Pricing / signals mentioned:

  • Quote range: $400K to $1M per project
  • Example: Lamini AI
    • Raised $25M
    • Example enterprise customer mentioned: AMD (Fortune 500)

Defensibility advantage:

  • Labeled “Sherlock-proof” because enterprise customers want sovereign AI on their own infrastructure/firewalls; platform model updates don’t replace that need.

Ranking: high barriers + high pay + longevity. S.

6) Data Labeling (RLHF “fuel” supply chain) — B+ tier for non-Scale; S+ for Scale

Core idea: Sell the training/feedback data needed to improve AI models—especially for RLHF (reinforcement learning from human feedback).

Process described:

  • AI generates candidate answers
  • Humans rank/grade responses
  • Feedback trains the model

Key company example:

  • Scale AI (Alexander Wang)
    • Positioned as a “toll booth” for AI development
    • Early labor model referenced: lower-wage image labeling in places like Kenya/India/Nigeria (“digitized sweatshops”)
    • Current labor model referenced: higher-skilled labeling (PhDs/math/engineering puzzles)
    • Outlier.ai referenced as associated

Capability & economics argument:

  • If you’re Scale AI, you win due to winner-take-all advantages (others want top-tier data).
  • Competitors are downgraded because it’s hard to match that data advantage.

Ranking logic: strong strategic position but hard to compete if you’re not the category leader. B+ (non-Scale) / S+ (Scale).

7) Vertical AI (build proprietary models for narrow domains) — A+ tier

Core idea: Build proprietary AI systems specialized for a specific industry/function rather than a general-purpose model.

Framework: reduce scope → become best at one thing.

Examples:

  • Harvey AI (elite law firms)
    • Claimed valuation: ~$8B
    • Claimed earnings: ~$200M/year
  • Midjourney
    • Claimed ~$500M annual revenue
    • Framed as proprietary modeling (“doesn’t pay OpenAI a penny”), implying a non-rental approach

Why it’s high-up (but hard):

  • Defensibility reversed: instead of fearing Sherlocking, vertical AI becomes the threat others fear.
  • High capital requirements:
    • Hiring 20+ PhDs
    • Buying ~$10M worth of GPUs (as stated)

Ranking: hardest to execute, strongest long-term positioning. A+.

Notable “playbook” themes / decision rules implied by the ranking

  • Defensibility vs. platform replication
    • Wrappers/agencies/influencers: vulnerable to platforms adding features or controlling distribution.
    • Enterprise consulting + vertical AI: more defensible due to private infrastructure/custom workflows.
  • Speed-to-market vs. durability
    • Programmatic SEO: monetizes quickly but collapses under algorithm enforcement.
  • Data as a strategic input
    • Scale/labeling positioned as essential “fuel” for model improvement.

Metrics & KPIs explicitly mentioned

AI wrapper examples

  • PhotoAI: ~$138,000 (historical)
  • Formula Bot: ~$226,000 MRR

AI automation agency economics

  • Zapier/Make: $20–$50/month
  • Client fees: $200–$2,000+/month (range described)

Programmatic SEO case

  • Jake Ward: 1,800 articles
  • Claimed traffic: 0 → 3.6M organic visitors
  • Arbitrage example: $1,000 spend → $50,000 ad revenue

AI influencer metrics

  • Karen AI: $71,000 in first weeks; $1/minute
  • Aitana Lopez: ~400K followers; “thousands/month in brand deals”

Enterprise AI

  • Pricing: $400K–$1M per project
  • Lamini AI: $25M raised
  • Enterprise customer example: AMD

Vertical AI

  • Harvey AI: ~$8B valuation, ~$200M/year
  • Midjourney: ~$500M annual revenue

General timeline mention

  • Programmatic SEO: “a few hours” to generate 1,800 articles; later “overnight” traffic collapse after Google action.

Presenters / sources mentioned

  • Pieter Levels (PhotoAI)
  • David Bressler (Formula Bot)
  • Liam Otley (AI automation agencies channel)
  • Jake Ward (programmatic SEO case)
  • Sam Altman / “ChatGPT” team (platform feature replication referenced)
  • Alexander Wang / Scale AI (data labeling)
  • Outlier.ai (referenced as owned/associated with Scale)
  • Lamini AI (enterprise AI consulting example)
  • AMD (Fortune 500 client example mentioned)
  • Harvey AI (vertical AI example)
  • Midjourney / David Holz (vertical AI example)
  • dot.online (sponsor)

Original video