Frontier models matter. Knowing your own house matters just as much.

Let me be clear up front: frontier models are a big deal. The reasoning, the coding, the long context. I follow every release and I'm not here to say any of that is overrated.

But I keep coming back to one idea. The AI winner may not be the one with the biggest breakthrough. It may be the one whose AI understands its own products, services, and customer workflows best. You need both. A brilliant model that doesn't know your business leaves a lot of value on the table.

Smart isn't the same as useful

Google is a good way to think about this, because it has both pieces. It has Gemini on one side, and a huge set of products on the other: Google Cloud, Workspace, Ads, Android, Maps.

Now imagine asking a general-purpose assistant why your Google Cloud deployment keeps failing. A strong model can explain permissions in theory. But can it look at how your project is actually set up and tell you which step in your pipeline is the problem? That's the gap between knowing about a product and knowing the product.

Or imagine a small business owner asking for help with an Ads campaign. A generic answer about keywords and budgets is easy. A useful answer knows how that person's account is structured, what they tried last month, and what they're really trying to sell.

These are made-up scenarios, but I think most of us have had some version of them. The model is clearly smart. It just doesn't know where it is.

Workflows are where the value hides

The real value is in the steps people actually take. How someone moves from a Gmail thread to a Doc to a calendar invite in Workspace. How an Android developer goes from a crash report to a fix. How a delivery business plans routes in Maps every morning.

An AI that understands those flows can do more than answer questions. It can act. That's why I like tools like Cursor's agent automations. You trigger an action and let your workflows run while you sleep. The model matters, but what makes it work is that it's built into how I already work.

The same goes for the bigger picture. Origin, Cloud Agents, the xAI API, and Grok Bots are quietly forming an end-to-end ecosystem. The pieces get stronger as they understand each other.

Your context should travel with you

One more thing, from the user's side. If context is what makes AI valuable, then the context we build up belongs to us. Continuity shouldn't end when you switch AI providers. Every model should let you export a complete repository of your prompts, ready to become context for your next model.

That would actually push companies toward what I'm describing. If you can't lock people in, you have to win by understanding their work better than anyone else.

So yes, keep building bigger and better models. But the companies that get the most out of them will be the ones that teach those models their own products, their own services, and the way their customers really work. Google has every piece it needs to do that. The question is who connects them best.