AI Tools: The 9 Stages of AI Agent Adoption, Explained

Watch on YouTube: https://www.youtube.com/watch?v=nRl-7Yhmbog

Channel: Greg Isenberg

Introduction: The Emerging Conflict Over AI Agents

Greg Isenberg frames the current moment as a “war on AI agents” — a struggle between platforms that block autonomous buying agents and those that welcome them. The stakes are high: agents act as “customers with credit cards and no eyes,” reading structured data instead of visual interfaces. Isenberg, a founder and advisor to companies like TikTok and Reddit, outlines a nine-stage adoption curve that moves from outright blocking to eventual normalization, with a critical “Plaid moment” for agent identity and payments in the middle. This post breaks down the framework, the numbers behind it, and what businesses should do today.

What the Video Covers

Isenberg identifies three camps shaping agent access:

• Blockers — Amazon and Delta actively prevent outside agents to protect high-margin upsells (ads, seat upgrades, baggage fees) and avoid disintermediation by Meta or OpenAI.
• Accidental blockers — Walmart and Zillow intend to allow agents but inadvertently block them through aggressive bot protection (WAF rules, Cloudflare challenges).
• Welcomers — Meta, Shopify, Visa/Mastercard, and Cloudflare are building infrastructure for agentic commerce, including agentic storefronts and a trusted identity/payment layer.

Agents ignore visual design, banners, and pop-ups. They read prices, specs, policies, and compatibility data in plain text. Their loyalty goes to whatever worked last time — reliability and speed, not brand nostalgia. The “ladder of trust” defines three rungs: research (widely allowed), prepare (cart/quote building), and pay (where the war lives).

Step-by-Step: The 9 Stages of Adoption

Isenberg’s speculative model traces a progression from conflict to routine:

1. Blocking — Incumbents erect walls (Amazon blocks Meta’s Muse agent; Delta allows zero outside agents for booking).
2. Arms race — Agent developers build evasion tactics; defenders tighten bot detection.
3. Market split — Welcoming platforms gain agent traffic; blockers lose transaction volume.
4. Trusted middle layer (Plaid moment) — Visa/Mastercard + Cloudflare create verified agent identity and payment rails, analogous to Plaid’s bank connectivity.
5. Welcoming companies win — Merchants on Shopify, advertisers on Meta capture agent-driven sales.
6. Agent competition — Multiple agents vie for the same purchase; speed and data quality become differentiators.
7. Agent SEO — Businesses optimize structured data (price, delivery, returns, compatibility) for agent readability.
8. Agent ads — Platforms like ChatGPT test in-chat ads and checkout (Etsy, Shopify); Gemini builds agents while protecting search ad revenue.
9. Normalization — Agent purchasing becomes as standard as web checkout today.

Tools, Numbers, and Named Entities

• Amazon: tens of billions per year in ad revenue at risk; blocked Meta’s Muse agent circa September.
• Delta: “zero outside agents” policy for flight booking.
• Yelp: agents allowed only via paid data licensing.
• Walmart: Muse partner but accidental blocks via bot protection.
• Shopify: rolling out agentic storefronts for merchants.
• Visa/Mastercard + Cloudflare: building the trusted agent identity/payment layer.
• Plaid: referenced as the model for agent-business connectivity (valuation in billions).
• ChatGPT: testing ads inside AI search; Etsy/Shopify checkout in-chat.
• Google Gemini: developing agents and payments while safeguarding search ads.
• ideabrowser.com: mentioned as a source for startup ideas in this space.
• $50: example spend limit for agent autonomy before human confirmation.

Caveats and Hype Check

Isenberg’s “9 stages” is a proprietary framework, not an industry standard. The “war” narrative amplifies tension for effect. While his track record (TikTok, Reddit, LCA) lends credibility, the startup ideas — vertical agent readiness consulting, Plaid-for-verticals (contractor booking, vet scheduling), agent mistake insurance, agent-traffic analytics, specialist suppliers — are plausible but unproven. No affiliate links or undisclosed affiliations (e.g., with Muse) were detected.

An agent is a customer with a credit card and no eyes.

Agents are loyal to whatever worked last time.

The customer has no eyes now. It’s already shopping. It’s reading instead of looking.

Are you protecting the customer or are you protecting a habit you’d like the customer to keep?

A business that can’t be bought by an agent is going to look like a business without a website looked in 2005.

Concrete Next Actions

Isenberg recommends five immediate steps to prepare for agent traffic:

• Test with real buying questions. Pull 10 questions from your support inbox, give each to 2–3 agents (ChatGPT, Claude, Perplexity, etc.), and document where they get stuck.
• Audit bot protection. Review WAF rules (Cloudflare, Akamai, etc.) — many block agents accidentally. Whitelist known agent user-agents intentionally.
• Expose critical purchase data in plain text. Put total price, delivery date, compatibility matrix, and return policy directly in HTML (not buried in JavaScript or images).
• Make competitive advantages machine-readable. If you offer next-day delivery vs. a competitor’s three weeks, encode that in structured data (schema.org, JSON-LD), not just marketing copy.
• Define permissions per trust rung. Allow research freely; allow cart/quote building; require human confirmation before payment on high-value orders (e.g., above a $50 autonomy limit).

Businesses that treat agents as a new customer segment — readable, permissioned, and measurable — will capture the early wave of autonomous commerce. Those that don’t risk becoming invisible to a growing class of buyers that never “visit” the site at all.