Michał Miszczyszyn, CTO and co-founder of Your Next Store, shared how the team is building an AI storefront builder for commerce that goes beyond generating polished mockups. Instead of producing static pages that break once real business logic enters the picture, Your Next Store creates storefronts connected to actual commerce flows, including products, cart, checkout, and payments. The talk focused on the technical architecture behind that system, from creating a sandbox for each session, to cloning a Next.js codebase, starting a dev server, running an AI agent over WebSockets, and updating the live storefront as the user iterates in natural language.

He also walked through the engineering trade-offs that appear once the product starts working end to end. Scope has to stay tightly bounded so the agent operates within an e-commerce context instead of trying to do everything. Prompt changes can have an outsized effect on output quality, latency, and cost, which makes iteration expensive and hard to predict. Sandbox lifecycle management becomes a product problem of its own, because teams need to balance fast resurrection and a smooth user experience with the cost of keeping sandboxes alive. He also pointed to security as an unresolved challenge, especially when the application code and the AI agent server live in the same sandbox and the system needs safeguards to prevent the agent from modifying sensitive parts of its own environment.

What you’ll walk away with:

• Why many AI website builders fail when they meet real commerce requirements
• How Your Next Store uses Vercel Sandbox, Claude, WebSockets, and Next.js to power live storefront generation
• Why bounded scope matters when designing AI agents for a specific domain
• How sandbox lifecycle management affects both UX and infrastructure cost
• Why security remains an open problem when AI agents can operate close to application code

Watch the recording to see how AI-generated storefronts can move from mockups to real commerce flows.