The Flippening
I had to buy a camera lens recently. Not through an AI. Through Amazon.
I opened a browser, navigated to amazon.com, logged in, went to my order history, scrolled until I found the last lens I bought to confirm the mount type, went back to search, typed a query, filtered by prime shipping, read four product pages with conflicting spec tables, checked a compatibility thread on a photography forum, went back to Amazon, added the item to cart, entered my address for the fifth time this year, and confirmed the order.
This took about twenty minutes. It felt like banging rocks together.
The contrast with what I now do routinely with AI was visceral. Not abstract-future visceral. Present-tense, this-week visceral. For anything that lives in language — research, drafting, analysis, comparison, decision support — the gap between the agentic AI experience and the old web-of-forms experience has already become almost offensive. The old way doesn’t feel neutral anymore. It feels like a punishment.
What’s still missing is the agentic layer on top of real-world action: the AI that doesn’t just advise on which lens to buy but goes and buys it, checking my order history autonomously, confirming the mount, cross-referencing my budget, and placing the order. That part is coming. It is not here at scale yet. But the infrastructure — the APIs, the model capabilities, the trust frameworks slowly being assembled — is being laid right now, mostly out of public view.
When the agentic layer arrives in force, the disruption will not look like previous tech transitions. It won’t be a gradual migration from desktop to mobile, or from physical to digital retail, where incumbents had years to adapt and users had years to adjust. The agentic transition will feel sudden because the prerequisite — models capable enough to act reliably in the world on your behalf — crossed a threshold quietly, and most of the ecosystem hasn’t processed it yet.
Consider what actually dies. Not Amazon. Amazon survives because it owns the fulfillment layer, the physical infrastructure, the last mile. But the Amazon interface — the search bar, the filter rail, the product page, the checkout funnel — becomes vestigial. If an AI agent can query Amazon’s inventory, confirm compatibility, apply my payment credentials, and place an order without a human ever loading a webpage, then the entire front-end edifice of e-commerce is no longer a product. It’s overhead. The same logic applies to airline booking, insurance forms, government portals, HR systems, healthcare scheduling, and roughly 80 percent of what people mean when they say they “use the internet.”
The companies that understand this are quietly building agent-accessible APIs and agentic interfaces. The companies that don’t are still A/B testing button colors.
The flippening — the moment when AI-mediated action becomes the dominant mode of getting things done in the world, and clicking around websites becomes the legacy fallback for edge cases — is still in the future. But the gap between what AI already does and what legacy interfaces demand is widening fast, and every time you feel the friction of the old way, you’re feeling the seam. The distance between that friction and intolerance of it is shorter than most people think.
The lens arrived in two days. The twenty minutes are not coming back.