We’re seeing AI features pop up in every product we use. Slack, Google Drive, etc.

Does this seem silly or useless to you? Me too. This is the last, desperate gasp of the Product Era in software design as we move into the Intelligence Era over the next couple of years. They’re pretending that “AI magic” will help their product continue succeeding and retaining users as AI gets more and more usage and mindshare.

The Product Era was defined by dedicated-purpose products protected by network-effect moats. I started using Google Docs in the mid-late ’00s because it had the best UX: it was just in the browser, and collaboration worked magically well. Writely was launched in 2005 and by 2006 it was acquired by Google and became Docs. Now there are lots of competitors with better features and product-specific UX, but Google keeps its users via its all-in product experience moat of “everything’s in Drive, everyone already has google accounts, and it’s annoying to use anything else to collaborate”.

Those network effects are strong—strong enough to make a big contribution to Google being the trillion-dollar company it is—but they will soon lose out to a new AI trend: the gains from integration with new AI models. Data and APIs will be stronger drivers of usage than UX. That’s why I call it the Intelligence Era, and I believe we’re about to see a fast transition.

For example: We’ve seen new companies who have trained a powerful new AI model to write code. Copilot, Cursor, Claude Code. Today, they distribute the coding features within the context of an existing app frontend—VS Code extension, VS Code fork, or Terminal app. But this is obviously not the end state. The existing user-facing product is just a vehicle to access a model, and it’s the model that matters for the user and what makes money for the developer.

The winners will be the model builders who make the most useful models! People will jump through hoops to get the best models. But model builders won’t always be the best product designers. That’s why I think the winning frontend will be the one that lets the user have the richest access to the most useful winning models from anywhere on the internet.

This is the opposite of the network-effect moat, which is why the product companies are all flailing. Google makes money by essentially discouraging you from sending your Google Doc data elsewhere. But the best new models are probably not going to come from within Google: e.g., someone’s going to make a really good fluent dictation partner—you just talk, and it types, asks you questions, and edits as you go. Someone else is going to make live editable translations of a doc, useful for cross-language collabs. Another startup will find citations to fact check you as you think. Some extremely basic app frontend will win not because it has the most features or the best ux, but because it’s best at letting you send your data out to models, enabling you to adapt your workflows quickly as new tools arise to match your needs.

I am not sure whether Google is capable or motivated to turn Docs into such a future frontend. Probably they are too constrained by existing Docs users and the network effects to do so.

I believe this threat extends to all the big productivity tools today that have enjoyed a reign of dominance due to having good network effects or good UX, but definitely it’s coming for text first (and probably images). Google Docs and Sheets, Notion, Photoshop, VS Code, and so on.

On the flip side: the future will come at us fast when we decouple product features from frontend. AI can build integrations with new APIs quickly on its own, so I can easily imagine an open source app frontend which Claude knows how to customize. When you need a feature like simultaneous translation of everything you type, you find an AI service that can handle it, then tell Claude to build its api into your frontend, and within a couple of minutes you have it working. This doesn’t really work to do today because frontend apps are quite complicated and the AI can’t reliably make changes like that, but as the frontends get simpler and Claude gets better, it will quickly start being possible and then easy.