What Google Did To Websites Is Happening To Your App Right Now
Google spent the last year turning websites into data sources. In the span of one week, 4 of the largest platforms on earth did the same thing to apps. Apple closes the loop on Monday.
We do not have to guess what happens to applications and SaaS. We have a finished case study on the coming commoditization cycle called the open web. For two decades, the deal was you published to a website, Google indexed it, Google sent you traffic, and you converted that traffic into ad revenue, subscriptions, or leads. The website monetized the destination, and Google monetized the roadmap. It’s the same relationship apps have with operating systems and super platforms today.
Then Google stopped being a roadmap and became a destination. AI Overviews support the user’s intent, and the user gets what they came for without leaving the results page. Ahrefs put the click-through decline for top-ranking pages near 58%. Pew found that when an AI Overview is present, roughly 8% of users click a traditional result vs 15% when it is absent.
Zero-click searches moved from 56% of queries to 69% in a single year. News organic traffic fell from a 2.3 billion monthly peak in mid-2024 to under 1.7 billion by the following spring. Penske Media is suing, and its filing explains the mechanics: Google went from a search engine that sends traffic to websites to an answer engine that removes the reason to visit one.
That mechanism generalizes to most AI action surface paradigms. Wherever workflows begin is taking over more of the workflow. The website did not disappear. Its content was more valuable than ever because the content is what the answer is built from. What disappeared was the website as the place you go. The content got extracted, and the container got bypassed. The data lived, but now it lives inside someone else’s interface, and that’s where the money is.
Now watch it happen to apps.
The Action Surface Replayed 3 Times In One Week
An action surface is the layer that sits above applications, takes a user’s intent, and completes the work by reaching into the app’s backend. The app stops being the place you go and becomes a thing the surface calls. The same disintermediation Google ran on websites, with the road becoming the destination.
In the last seven days, three platforms announced it in three different markets.
Microsoft, at Build 2026, turned Windows into an agent runtime. The framing from the keynote floor was that Windows is no longer a platform for running applications. It is becoming a runtime for registering and orchestrating agents that take action across them. Apps register an agent, the OS passes it context, and the agent acts. The user doesn’t see what’s happening behind the action surface.
Meta took WhatsApp’s business layer agentic. The Meta Business Agent went global yesterday across WhatsApp and Instagram DMs. Underneath it sits the Business Agent Platform, which plugs into hundreds of backend systems (Shopify, Zendesk, etc.) and takes action on the business’s behalf. WhatsApp was already the commerce layer for the informal economy across the Global South. Meta just put an agent on top of it. The boutique owner’s app is now a row of capabilities the agent invokes inside a chat thread.
Tencent is making WeChat the action surface for 1.4 billion people. ClawBot landed in March. The native agent in development is designed to orchestrate roughly 3.8 million mini-programs. The internal goal is to convert WeChat from a social entry point into an intelligent task center where you state intent and the platform executes across the ecosystem. Mini-programs were already apps-without-installs. Now they are functions the surface calls.
Three platforms and markets with one architecture. The interface gets annexed by a layer none of these app developers control.
Those Were The Warnings & Apple Is The Event
Windows owns the desktop, WhatsApp owns messaging commerce, and WeChat owns China. Apple owns the highest-value consumer install base on the planet, and it might make its move on Monday.
WWDC 2026 opens June 8. iOS 27 brings the revamped, Gemini-backed Siri, and the piece to watch is App Intents. Intents is the mechanism that lets Siri operate third-party apps the way a person would. Find a photo, edit it, and send it. The spring release already shipped on-screen awareness and basic cross-app actions. iOS 27 could turn it into the full conversational layer that drives apps by intent.
Apple is the inflection and not just another entry point. AI Overviews were not dangerous because the technology was new and cool. That doesn’t drive adoption. They were an extinction event because Google had the distribution to make them the default for billions of people at once.
The capability had been demonstrable for a while, but the scale is what converted it to an event. Apple has that kind of scale for consumer apps. When Siri becomes the surface, and your app becomes a set of intents it calls, the consumer half of the market converts on Apple’s timeline, not yours. The phone becomes the action surface, and the app becomes a callable commodity.
Compression Into The Middle
Any application can be deconstructed into three layers.
At the top is the interface: the screens, navigation, brand, and relationship with the user.
In the middle is the logic (information layer): The workflow, business rules, and orchestration that the UI exposes step by step.
At the bottom is the data layer: The records, schema, and state.
The action surface annexes the top and becomes the interface. The user talks to the surface, not your app. The system of record claims the bottom, and the emerging semantic layers are all vying for the middle. The data and information are the valuable assets, and everyone knows it. What gets squeezed is the middle, and the middle was the product when logic, information, and data access lived in code.
This is the uncomfortable part for most software companies. The UI was never the value. It was where you charged monthly for access to the logic and the data behind it through seats, training, and the switching cost of making humans re-learn a different set of screens. Remove the requirement that a human navigate your screens, and the workflow underneath turns out to be a handful of function calls. The UI comes down, and the road behind it was shorter than anyone admitted.
That compression starts a self-reinforcing commoditization cycle. To stay reachable on the surface, you must expose your intents and APIs to it. But exposing them strips your differentiation. No more screens to make memorable, brand the user sees, direct relationship, or experience to compete on. You are reduced to the quality of your function calls, and function calls are easy to swap.
The more you integrate to survive, the more interchangeable you become. The more interchangeable you become, the more the surface arbitrages you against the next interchangeable backend. The race to expose yourself is a race to the bottom on margin. Both enterprise and consumer apps get pushed toward the same thin shape, for the same reason.
Most Enterprise Software Is A Database Wearing A UI
Strip away the interface from a large share of enterprise SaaS and what remains is a schema, some validation rules, a workflow, and an API. It is a database with a UI to make it useful to people. The SaaS moat was never the logic. It was switching costs, user training, and integration friction, all of which lived in the interface layer.
They were artifacts of the fact that humans had to operate the software. Agents do not need to be trained on an app’s screens. Nor do they feel switching cost. The friction SaaS companies monetized was friction the action surface removes by design.
There is a clear line between what survives this and what does not. Software that computes something hard (a pricing engine, risk model, optimizer, simulation, or anything where the logic is the genuinely difficult and defensible part) keeps its value because that value was never in the UI to begin with.
Software that stores and displays something gets compressed to its data layer, because that is all that was ever valuable. That’s the majority of CRMs, ERPs, and enterprise productivity software.
The Databases Want To Be 1 Database
Here is the second-order move, and it is where the next decade of enterprise architecture gets decided. Agents are worse than humans at operating across fragmented systems of record, and most enterprises are nothing but fragmented systems of record. Two hundred SaaS applications, each with a private database, schema, dialect of API, copy of ‘the customer’ or ‘the product’, and version of truth.
The UI era hid that fragmentation behind human labor. People reconciled, re-keyed, and cross-referenced between systems all day. That work was invisible because people accepted and managed it.
An agent inherits all of that fragmentation with none of the human domain expertise and adaptability. To act, it must resolve which of six systems holds the authoritative record, stitch identity across stores that disagree, and reason over conflicting states. Agents pay the full integration tax on every action instead of it being hidden by a department. An agent operates most efficiently with one system of record with consistent semantics it can read and write against, ideally transactionally.
The economic pressure runs toward consolidation. The databases that survive the great UI unwrapping do not survive as two hundred islands. They collapse into a far smaller number of systems of record, or get unified under a single semantic layer (knowledge graph) that the agents query. That’s what Databricks, Snowflake, Microsoft, and many others are fighting over. It’s why Microsoft cut access to Databricks agents when it tried to bypass Fabric’s semantic layer.
The fragmented application landscape was an artifact of human interfaces. Every app got its own database because every app needed its own screens for its users. Take the screens away, and there is no reason for the data to stay shattered into hundreds of pieces, and a strong reason for it not to. The agent landscape rewards consolidation the way the human landscape enabled fragmentation.
What For SaaS & Apps
The websites that survived Google’s pivot did it one of two ways. They owned a direct relationship that did not route through search. Future plc built an entire strategy around ‘Google Zero’. The survivors have leaned into subscriptions, communities, and reasons to be the destination, or they produced something the answer engine could not synthesize away.
The apps that survive the action surface have the same two doors, plus a very limited third.
SaaS companies can try to be the surface, but four companies with billions of users each have already claimed those seats, and they probably are not getting one.
SaaS companies can be the hard logic or the computation valuable enough that the surface calls it as a premium function rather than replacing it. That requires products to be hard in a way that most enterprise software is not.
Or SaaS companies can be the system of record the agents consolidate onto: the canonical place the data lives and the layer everything else reconciles to (data and eventually logic/information).
Apps without one of those chairs are in the middle. The middle is a database wearing a UI, and the UI is being taken away. The website collapse quantified how this ends. The only thing SaaS companies and apps got was an extra year to read the case study before the same thing happens to them.
The question SaaS companies must answer is which of those three things it really is. If the honest answer is none of them, the surface will soon replace you.
I teach leading-edge product strategy concepts like these in my AI Product Management Certification course. My next cohort starts on July 11th. Space is limited, so act fast to secure your space.