Ask HN: How do we build an "Agent Gateway" for the 1.5B legacy websites?

Anh_Nguyen_vn1 pts0 comments

I m a solo founder from Vietnam building in public. Looking at recent data, 57.5% of internet traffic is already automated (bots, LLM scrapers, custom frameworks). The other 42.5% doesn t know what s coming.The core problem: Most existing websites (WordPress, Shopify, legacy setups) were built for human eyes, not AI agents. Rebuilding them from scratch to be machine-readable is financially and technically impossible for most small businesses.With Anthropic’s MCP crossing millions of downloads and HTTP 402 (Payment Required) becoming real at the network edge via Cloudflare/Coinbase/Stripe, the infrastructure for an agentic web is converging fast. AWS and Cloudflare are shipping tools, but they are built for enterprise engineering teams, not the long tail of the web.I envision an Agent Gateway —a thin reverse proxy layer that sits in front of a legacy website without changing the frontend: - Human visits - normal HTML website - AI Agent visits - gateway routes to an agent tunnel - serves auto-generated structured JSON, llms.txt, agent.json, or an MCP endpoint. - Incorporates x402 micropayments at the edge.As a non-systems architect, I m stuck on a few engineering questions: 1. Session State: How do we handle authentication for autonomous agents that don t retain cookies or traditional session states? 2. Abuse Prevention: Since this removes CAPTCHAs, does rate-limiting shift entirely to x402, or do we need behavior-based fingerprinting for agents? 3. Liability: If an agent acts on stale catalog data cached by the gateway, who holds the liability?I d genuinely love to hear how engineers here would architect this. Is a Cloudflare Worker proxy the right approach, or something else entirely?

agent gateway legacy agents cloudflare websites

Related Articles