Cloudflare will make 1000+ AI millionaires

The old paradigm: crawl, visit, monetize

0:00 Isenberg frames the episode around three questions: what Cloudflare is actually doing (in plain English), why it creates a new business model for the internet, and which startups to build on top of it.

1:20 The old bargain: a website let search engines crawl it because crawling sent traffic. Google indexed the site, a human searched, the page showed up, the human clicked through, and the site monetized that visit (ads, email capture, a subscription, an affiliate link). The crawler got the content, the site got the visitor, the visitor got the answer, and that funded a large part of the internet.

An AI agent breaks this loopit can read a page, pull the useful answer, and hand that answer to the user - and the user may never visit the site. The content still created value, but the site can lose the visit, the ad impression, the email capture, and the affiliate click.

2:52 The reframe: the internet needs a pricing model for machine usage. A human won't pay a fraction of a cent to read a recipe, but a machine doesn't care if a payment is tiny and automatic - if the data helps an agent finish its job, the agent can just pay. Isenberg's mental model: the human web monetized attention; the agent web monetizes useful resources.

What Cloudflare is actually doing

3:41 Four pieces, in order:

• AI Crawl Control - gives site owners visibility into which AI crawlers are hitting their content, and lets them allow or block specific ones.
• Pay Per Crawl - the direct monetization piece. A site owner can charge AI crawlers for access; the crawler either presents payment intent up front or gets a 402 Payment Required response with pricing.
• Monetization Gateway - the broader version. Any Cloudflare-protected resource (a page, a dataset, an API, an MCP tool call, a premium endpoint, a search index) can carry payment rules, not just crawler-read pages.
• x402 - the payment rail underneath both, built on the (previously unused) HTTP 402 Payment Required status code: the agent requests a resource, the server states a price (often a fraction of a cent), the agent pays and retries with proof of payment, and Cloudflare verifies at the edge before the request reaches the origin.

4:57 The bigger idea, in Isenberg's words: Cloudflare is trying to make paid access feel like part of the internet itself - no giant checkout flow, no account creation, no sales call, no enterprise procurement. The request becomes the transaction. That means small resources on the internet - a dataset charging per lookup, an API charging per successful call, a research archive charging per answer, a product catalog charging per comparison - can become standalone businesses. His anticipated objection ("agents don't have wallets") he waves off: agents are already gaining wallets and email addresses, and transactions are starting to follow.

6:29 Cloudflare's AI Index points the same direction: making sites easier for AI systems to use via MCP, LLMs.txt, search APIs, and bulk data APIs. Humans tolerate messy sites (they click around, open a 2019 FAQ, suffer through a PDF); agents need clean doors. The one-line shift: websites become resource layers, content becomes indexes, expertise becomes structured/callable, data becomes metered, and tools become agent-accessible.

The Agent Internet Stack

7:34 Isenberg's synthesis of the stack forming around this shift:

1. Messy internet - PDFs, pricing pages, old blog posts, YouTube videos, support docs, comparison sites.
2. Structuring - someone cleans that into structured data.
3. Agent-readable access - an API, MCP tool, search index, feed, or LLMs.txt exposes it to agents.
4. Payment rules - some resources stay free (distribution), some are paid (high value), some are blocked (private).
5. Trust and analytics - is the data fresh, is the source reliable, which agents are using it, which requests are worth money, which resources drive outcomes.

Why now is the best time to build

9:09 Isenberg addresses pessimism circulating on X: indie hacker Pieter Levels (@levelsio) posted that he's seeing declining revenue and traffic across indie-hacker projects, including his own, and pinned it on "BigAI... cannibalizing everything that used to be apps," quote-tweeting Jon Yongfook (@yongfook) whose own post showed Google referral traffic roughly halving from January to July 2026.

9:38 Isenberg's counter-take: this is the best time to build, but not for vibe-coded point tools - for startups with real moats, built for an AI-native, agent-ready internet, the way early App Store apps were built for a platform still forming (he draws the 2009 App Store analogy again in the closing section).

Startup Idea 1: The Niche Data Refinery

10:29 Pick one niche where valuable information is messy, fragmented, changing, and annoying to collect, and refine it into clean fuel for agents. The raw material already exists (Google Maps, job posts, reviews, local directories, PDFs, pricing pages) - the business is the refining.

Worked examplemed spas. An owner wants to know competitor pricing, treatments, review complaints, hiring signals, and market trends - scattered across Google reviews, competitor sites, Instagram, ad libraries, and employees' heads. An agent with that data cleanly assembled could say things like "your Botox pricing is above the local median but your reviews don't support premium positioning yet," or "three competitors started promoting exosome treatments in the last 60 days."

Build sequencepick one niche and one city -> track about 100 businesses manually -> build a spreadsheet (name, website, services, prices, review count/rating, top complaints, Instagram, recent posts, ad changes, hiring signals, booking flow) -> produce roughly 10 outputs from it (pricing map, competitor gap report, offer ideas, services-to-add list, review-complaint summary, hiring-signal report, monthly market-movement report).

The first customer usually isn't the end business - it's whoever already sells into that niche (marketing agencies, consultants, freelancers, software vendors, AI-implementation people). A med-spa marketing agency selling a $5K/month growth package can pay $300-800/month for data that helps it land or improve one more client. Crawl-walk-run: report -> dashboard -> API -> MCP tool -> agents paying per lookup once the rails exist.

Other verticals Isenberg names with their own data axes: roofing (storm events, permits, insurance signals, reviews, ad angles), real estate investing (zoning changes, permits, ownership records, rent comps, tax delinquencies, insurance shifts), e-commerce (competitor SKUs, pricing changes, review complaints, influencer rates, UGC hooks, Shopify apps, shipping promises), law firms (local competitors, practice-area positioning, ad copy, reviews, intake).

The filter for whether a niche qualifies: the data must be valuable (better decisions, money made or saved), repeated (needed again and again, not a one-off), changing (freshness matters), fragmented (no one person can easily collect it), and annoying (which is where the margin lives).

Startup Idea 2: Agent Readiness for Businesses

17:15 Framed as "SEO for the agent internet," though Isenberg is careful to distinguish it from the increasingly buzzy "AI SEO" label: the real business is helping companies become easy for agents to understand, trust, compare, and recommend.

A human B2B buyer works through a homepage, pricing page, docs, a demo, a case study, maybe a friend's opinion. An agent compresses all of that into one pass: who is this for, what does it cost, what does it replace, what integrations does it have, what are the risks, what do customers say, how does it compare. Most sites make this hard - hidden pricing, buried docs, stale comparison pages, key facts trapped in PDFs, and vague marketing copy ("unlocking operational excellence").

Wedgerun a paid audit. Pick one vertical (B2B SaaS is the obvious one, but also Shopify apps, law firms, health clinics, financial advisors, insurance brokers, home services), then run 20-50 buyer-intent prompts across major AI tools ("what's the best software for X," "compare this company to alternatives," "what does this cost," "who is this best for," "what are the risks," "would you recommend this for a 20-person company," "what integrations does it support"). Show the company the answers - this is the sales moment: you don't show up, or you show up with the wrong price, or a competitor gets recommended because their docs are cleaner, or the right answer is buried in a 2002-era PDF.

Then sell the fixa clean llms.txt file, better documentation structure, a pricing page agents can parse, honest and specific comparison pages, plain-language use-case pages, customer proof organized by segment, structured FAQs, schema markup, a product feed, a changelog, and - if there's enough content to justify it - a lightweight MCP server or search endpoint. The recurring product is the measurement loop: rerun the prompts monthly, track whether AI-answer accuracy improves, whether the company appears more often, and how competitor comparisons shift.

Pricing Isenberg suggests (his own opinion, not observed market data): $3,000-$10,000 for the audit and cleanup, $10,000-$20,000 for larger B2B companies, productizing into software after roughly 10 clients in the same niche reveal the same repeated gaps. The pitch line: "you're not selling the future, you're selling the screenshot."

Startup Idea 3: Expert Archives as Agent Tools

23:40 For creators, media companies, analysts, and consultants sitting on years of content (videos, podcasts, newsletters, templates, community posts) that today monetizes through ads, sponsorships, subscriptions, or consulting. In the agent internet, that archive can become a callable tool: a founder agent that critiques your idea using a specific archive's framework, a sales agent that rewrites a cold email using one trainer's method, a fitness agent that builds a plan from one coach's philosophy.

The key discipline is starting from one specific job, not "turn your whole brain into AI" (which Isenberg calls "a SaaS landing page that should be illegal"). "You have 300 videos about sales - we'll turn them into a tool that improves cold emails" beats a general chat-with-the-expert product.

Build sequencepick an expert with a deep archive and a specific audience -> collect the archive (transcribe video/podcasts, pull newsletters, clean docs) -> tag it by job, topic, audience, example, framework, and outcome (skipping this and dumping everything into a vector database gives you "a search box with confidence," not a real product) -> build one useful workflow (for a sales expert: paste a cold email, the agent critiques it using that expert's principles, cites the source lessons, rewrites it, scores it, and gives one test to run).

Isenberg cites his own product, ideabrowser.com, as a live example: its MCP feature is one of its most popular, letting an LLM query the platform's own cleaned startup-idea data directly.

Monetization options$19-99/month subscription, bundled into a paid community, used as a consulting lead magnet, or licensed to agencies/software companies. The creator already has distribution and trust; the agent tool lets them democratize their expertise without doing 1:1 consulting for everyone.

The filter for finding ideas

29:47 What connects all three ideas: agents need clean, trusted, useful resources to do good work, whether that resource is data, structure, access, expert knowledge, a tool, or a payment rule. Cloudflare is building part of the access-and-payment layer, but you don't have to wait for it to mature - build the manual version first, sell the human version now, and package it as agent payments become common.

30:36 The five-question filter for a new idea: what decision is expensive, what information is messy, what changes often, who already pays for help with it, and what would an agent need to do the job better. Isenberg's App-Store-2009 analogy returns here: the best opportunities look small now because the agent internet itself is still small relative to where it's headed.

32:33 Closing thought: "the internet is shifting from pages humans visit to resources agents use." Agents are becoming buyers, websites are becoming resources, and the next great internet businesses may be the tiny paid doors agents walk through all day.

Tools and resources

Claims to verify

1. Cloudflare feature names - VERIFIED. "AI Crawl Control," "Pay Per Crawl," and "Monetization Gateway" are all real, correctly-named Cloudflare features. The Monetization Gateway was announced 2026-07-01 as the next step up from Pay Per Crawl, extending per-access billing to web pages, APIs, datasets, and MCP tools. Source: https://blog.cloudflare.com/monetization-gateway/
2. x402's origin - the video's framing is incomplete. x402 is an open HTTP-402-based payment protocol originally authored and open-sourced by Coinbase in May 2025 (github.com/coinbase/x402), now under Linux Foundation governance backed by a 25+ member industry coalition (including Cloudflare, Google, Visa, Mastercard, Stripe, AWS, and Circle) via the x402 Foundation. The video presents x402 as part of "what Cloudflare is doing" without crediting Coinbase's origination. Cloudflare is a coalition co-developer, not the protocol's creator. Sources: https://github.com/coinbase/x402 and https://blog.cloudflare.com/monetization-gateway/
3. The 402 payment flow - VERIFIED. The described sequence (agent requests a resource -> server returns 402 Payment Required with a price -> agent pays and resubmits with proof -> a facilitator verifies -> the resource is delivered) matches both Cloudflare's and the x402 protocol's own documentation. Settlement is in stablecoins, peer-to-peer to the seller's wallet, with zero protocol fees. As of March 2026, public reporting put x402 at 119M+ transactions on Base and 35M on Solana (roughly $600M in annualized volume).
4. "An AI Index for all our customers" - VERIFIED. This is a real, correctly-titled Cloudflare Blog post (the video's chapter title is a direct quote), covering an auto-generated AI-optimized search index per customer domain with an MCP server, LLMs.txt, and a search API, owned and monetizable by the customer. Source: https://blog.cloudflare.com/an-ai-index-for-all-our-customers/
5. The Levels.io/Yongfook traffic-decline claim - visually confirmed as a real post, underlying cause not independently verified. The on-screen screenshot is a genuine @levelsio post (Jul 26) quote-tweeting @yongfook (Jul 25) with a real traffic chart. Whether "BigAI cannibalizing apps" is the actual cause of the decline (versus other factors) is Levels.io's own interpretation, not independently verified data.
6. Pricing figures for the "agent readiness" service ($3-10K audits, $10-20K for larger B2B) are Isenberg's own suggested pricing, not a cited market benchmark - treat as advice/opinion, not observed data.

Glossary

• AI agent: software that can take multi-step action toward a goal (browsing, calling tools, making requests) rather than just answering a single prompt.
• Crawler: an automated program that visits and reads web pages, historically to build search indexes.
• HTTP 402 Payment Required: a reserved HTTP status code meaning "you must pay to get this resource"; historically unused, now being revived for machine payments.
• x402: an open protocol built on the HTTP 402 status code that lets an automated client pay for a resource and retry the request with proof of payment, with no login or checkout flow.
• Pay Per Crawl: a Cloudflare feature letting a site owner charge AI crawlers per visit or access to its content.
• Monetization Gateway: a Cloudflare feature extending pay-per-access rules beyond crawlers to any resource behind Cloudflare (APIs, MCP tools, datasets, files, search indexes).
• AI Crawl Control: a Cloudflare feature giving site owners visibility into which AI crawlers are accessing their content and the ability to allow or block specific ones.
• AI Index: a Cloudflare initiative making sites easier for AI systems to use via structured, agent-readable formats (MCP, LLMs.txt, search/bulk APIs).
• MCP (Model Context Protocol): a standard letting an AI model connect to external tools and data sources in a structured way, so an agent can call a tool instead of scraping a page.
• LLMs.txt: a proposed file standard (like robots.txt) that tells AI systems what's on a site and how to use it cleanly.
• Agent-readable: content or a resource structured (via API, MCP, feed, etc.) so an AI agent can reliably parse and use it, as opposed to a page meant for human eyes.
• API (Application Programming Interface): a defined way for one piece of software to request data or actions from another.
• Vector database: a database that stores content as numerical representations of meaning so it can be searched by similarity rather than exact keywords; commonly used for AI search/retrieval.
• Schema markup: structured code added to a web page that tells search engines and AI systems what the content means (for example, "this is a product with this price").
• Procurement agent: an AI agent acting on behalf of a business buyer to evaluate and compare vendors before a purchase.
• Buyer-intent prompt: a search-style question ("what's the best X for Y") used to test what an AI assistant recommends in a given category.
• Wedge: the narrow, specific first offer a startup uses to land its first paying customers before expanding.
• Crawl-walk-run: a staged rollout strategy, starting manual/small (crawl), then semi-automated (walk), then fully automated/scaled (run).
• Moat: a durable competitive advantage that protects a business from being easily copied.
• Data refinery: Isenberg's term for a business that turns raw, messy, publicly available information into a clean, structured, sellable product.
• Productize: turning a custom service (a manual audit or report) into a repeatable, packaged software product.
• Archive-to-API: turning a body of existing content (videos, podcasts, docs) into a queryable tool other software can call.
• Cold email: an unsolicited sales or outreach email sent to someone with no prior relationship.
• SaaS (Software as a Service): software sold as an ongoing subscription, hosted by the vendor rather than installed locally.
• AI Overviews: Google's AI-generated summary answers shown above traditional search results, which can reduce clicks to the underlying source sites.
• Vibe coding: building software quickly with heavy AI assistance/generation rather than traditional hand-coding, often producing lower-durability results.
• Stablecoin: a cryptocurrency designed to hold a stable value (typically pegged to a currency like the US dollar), used here to settle x402 payments.
• Retainer: an ongoing monthly fee a client pays for continued access to a service, as opposed to a one-time project fee.