And Why Google's WebMCP Is the Wrong Answer
Amazon just proved that embeddable AI agents are worth $10 billion in incremental annual revenue.
Rufus—Amazon's shopping assistant—hit 250 million active users in under two years. Customers who engage with it are 60% more likely to complete a purchase. On Black Friday 2024, Rufus-assisted sessions converted at 3.5x the rate of non-assisted sessions.
This isn't a pilot. This isn't a projection. This is $12 billion in actual incremental sales for 2025.
The message is clear: conversational AI on your website isn't a nice-to-have. It's a revenue engine.
But here's what nobody's talking about: Amazon built Rufus over years, with thousands of engineers, trained on decades of catalog data, customer reviews, and behavioral signals. They spent an estimated $285 million just on operating costs in 2024 before it turned profitable.
What about the other 400 million websites that don't have Amazon's resources?
The $28 Billion Question
The AI shopping assistant market is projected to reach $28.54 billion by 2033. Conversational commerce is already valued at $8.8 billion and growing at 14.8% CAGR.
The demand is here. Shoppers now expect AI-first experiences:
- 95% of customer interactions are expected to be AI-powered by 2025
- 47% faster purchase completion when AI assists the journey
- 23-70% conversion improvement across industries using AI chatbots
- 35% of abandoned carts recovered through conversational re-engagement
But the supply side? It's broken.
The Current Options Are All Wrong
Option 1: Build It Yourself (The Amazon Path)
You need:
- RAG pipelines that require months of development
- Constant maintenance as your product catalog, pricing, and features change
- Vector databases, custom LLMs, and evaluation frameworks
- An entire ML ops team to prevent hallucinations
Cost: $500K-$2M+ annually. Timeline: 12-18 months to production.
Option 2: Intercom, Drift, or Traditional Chatbots
These tools excel at support ticket deflection. Intercom's Fin boasts 50% ticket reduction—impressive for support.
But they're fundamentally reactive. They wait for questions. They don't guide users through checkout flows. They don't show product demos. They don't click buttons or fill forms.
And the setup? You're still building RAG pipelines, hiring consultants, and paying $0.99 per resolution. Every. Single. Time.
Option 3: Google's WebMCP (The New Trojan Horse)
Two days ago, Google announced WebMCP—a protocol that asks websites to expose their internal APIs as "tools" for Chrome's AI agent.
The pitch sounds helpful: "Make your site agent-ready and we'll make interactions faster!"
Sound familiar?
This is the same playbook as Google AMP, featured snippets, and News aggregation. You do the integration work. You maintain the schemas. You expose your APIs.
And Google's agent becomes the interface your users talk to—inside Chrome, not inside your site.
WebMCP says: Build tools so Google's agent can use your site.
Your checkout flow. Your onboarding. Your conversion funnel. All intermediated by someone else's agent.
The Alternative: Own Your Agent
We built Rover to solve a different problem: What if every website could have an Amazon-quality AI agent, deployed in one line of code?
No RAG pipelines. No API exposure. No schema maintenance. No handing Google the keys.
Rover is a DOM-native AI agent that lives on your website. It reads your actual page structure—buttons, forms, navigation, content—and understands it semantically. No screenshots. No scraping. No vision models guessing at pixel coordinates.
One <script> tag. Your agent. Your site. Your users.
What Rover Actually Does
- Completes checkout flows — Clicks buttons, fills forms, handles multi-step processes
- Onboards users in real-time — Interactive product tours that adapt to user behavior
- Shows live product demos — No more stagnant videos; actual walkthroughs of your product
- Builds complex workflows — Guides users through sophisticated processes step-by-step
- Converts conversations into actions — Users ask, Rover does
This isn't a chatbot. It's an agent that understands your website at the DOM level and acts on behalf of your users.
Why DOM-Native Changes Everything
Most AI web agents use one of two approaches:
Screenshot-based agents (OpenAI Operator, Computer Use, etc.) take pictures of your website and use vision models to guess where to click. They're slow, expensive, and fail on dynamic content.
API-dependent solutions (WebMCP, custom integrations) require you to expose every action as a structured tool. You do the work. You maintain it forever.
Rover is different.
We built Smart DOM Trees—a semantic understanding layer that maps your website's actual structure. Rover knows that "Add to Cart" is a purchasing action, that the sidebar is navigation, that the modal is a form.
This is why we're ranked #1 on WebBench with an 81.39% success rate—higher than any screenshot-based approach.
And it's why 21,000+ users have executed 1.5M+ workflows on our platform.
The Math for Websites
Let's be conservative:
| Metric | Without Agent | With Agent (Conservative) |
|---|---|---|
| Conversion Rate | 1.65% | 2.03% (+23%) |
| Cart Abandonment Recovery | 0% | 12% of abandoned |
| User Onboarding Completion | 40% | 58% (+45%) |
| Support Ticket Volume | Baseline | -35% |
If Amazon's Rufus drives $12 billion on a $600 billion GMV platform (~2% lift), what could a 23% conversion improvement do for your business?
For a site doing $10M annually, even a 15% lift is $1.5M in incremental revenue.
Rover costs less than a single support hire.
Rover Is Live
We've been building toward this moment for two years. DOM-native architecture. Smart DOM Trees. State-of-the-art web page understanding.
Rover is now live on rtrvr.ai. You can try it yourself.
And on February 25th, we're going live with our first waitlist companies on Product Hunt.
If you want:
- Your own AI agent that works for your users (not Google's)
- One-line deployment with zero RAG configuration
- Conversion rates that match what Amazon sees with Rufus
- Control over your checkout, onboarding, and user relationships
The Battle Lines Are Drawn
Websites already lost the content layer to aggregators. Google took search. Social platforms took distribution. AI answer engines are taking queries.
Now the question is: Who owns the interaction layer?
Google's WebMCP is a bid to intermediate every transaction, every onboarding flow, every conversion funnel—through Chrome's agent.
We're building the opposite.
Rover puts the agent on your turf. Your website. Your users. Your data. Your revenue.
The future of the web isn't websites becoming backends for someone else's AI.
The future is every website having its own intelligent agent.
That future is Rover.
rtrvr.ai is the DOM-native AI web agent platform, built by two ex-Google engineers. We're ranked #1 on WebBench, with 21,000+ users and 1.5M+ executed workflows.
Rover launches February 25, 2026 on Product Hunt.