AI-Native Apps Will Swallow the Web
Why we're publishing eight new MCP guides
Here is my prediction: AI-native apps will swallow the web. Not in ten years, but before 2026 is over – and the transition is already underway.
MCP – the Model Context Protocol Anthropic introduced in November 2024 – has already been adopted by Google, Microsoft, and the W3C. Every major LLM speaks it. The browser vendors are integrating it directly: WebMCP is a draft at the W3C with an early implementation already shipped in Chrome. The naming is still loose – agentic apps, agent-native apps, in-chat applications – but the category is real.
MCP is going to swallow the web just as the web did the desktop. AI-native apps (or whatever they come to be called) are the next web.
The web we have will not vanish. It will be subsumed. Before 2026 is over, AI-native-only will be a viable path: a product that exists as MCP-callable tools and machine-readable state, with no browser, no mobile app, no human-facing surface – and a real customer base. The agents are already there. The protocol is in place.
Users will pull this forward because the AI becomes both the access point and the source of enrichment. One conversation instead of a dozen tabs. The AI knows the sneakers, knows the cars, knows how you like to travel. Like the whole of the web, the merchant’s site becomes raw material.
This sounds extreme, but it is also already underway. Shopify has built a Storefront MCP for AI-mediated commerce. AI-native apps are bringing back the constraints that shaped the early web – stateless protocols, server-side logic, forms for input – for a new class of consumer.
It is because of this seismic shift that our client work has reorganized around it, and it is why we just published a new cluster of MCP guides – eight pieces written for product leaders deciding whether and how to be present inside leading AI clients. We start with the working vocabulary the category has not yet settled, lay out a strategic decision framework that determines posture, set the client comparison matrix for choosing where to ship first, walk through the embedding-types breakdown that distinguishes read-only from actions from agent-resident, frame the auth and security expectations enterprise procurement actually wants, work the build-vs-buy economics of in-house engineering versus a development partner, give you a partner evaluation checklist for a category too young for the usual portfolio signals, and tie it together in a hub guide. The category is moving fast enough that vague advice ages quickly. These guides are specific, current, and built to be useful for the decisions our clients are making right now.
AI-native apps will swallow the web.
