Ask Paul: July 10 ⭐

Ask Paul: July 10
Happy Friday! I was expecting Xbox existential dread this week and was happily surprised it didn't happen. But here's a great set of reader questions to keep our minds off those troubles and get the weekend started a little early.
"Layoffs begin at Ubisoft following Assassin's Creed Black Flag Resynced launch"

Thanks for helping us get that out the door, guys. Now show yourself the door
🔮 When will then be now?
NickTech asks:

Paul, what are your thoughts on the news that Microsoft is developing its own AI LLMs? Do you think Microsoft can put out something that is just as good (or even better) than the current LLMs in use, or could this cause people to migrate off Copilot?

Microsoft has been working on its own LLMs for years, but the drive to create in-house "frontier" models (as we now call them) that can compete with the best took off when Microsoft and OpenAI reconfigured their partnership earlier this year, giving them (Microsoft) the ability to pursue super-intelligence without OpenAI's help or involvement.

The news this past week is that Microsoft has been transitioning to its own models more than we previously knew, though as I wrote that story, I realized that it was less dramatic than the headline of the original report suggested. Microsoft has been open about this transition since Build 2026 in May, when it announced several new top-tier Microsoft AI (MAI) models.

Building AI models that can compete with the best in the market can't happen overnight, even for Microsoft, which has been using and enhancing OpenAI's best models for over three years. In some ways, this situation is a bit like the component crisis, where even the most well-equipped silicon makers can't just spin up a new memory fabrication facility and solve the problem quickly. But this is software, and Microsoft does have the expertise and, as crucially, the infrastructure to pull it off. So I feel like it's inevitable that Microsoft is competitive in this way.

But competitive is somewhat subjective and varies almost by day because new model releases always reset the rankings across whatever various workloads these models are measured against (coding, reasoning, generation, etc.). Gemini may or may not be competitive today compared to the latest and best models from Anthropic and OpenAI on any given day or week, but the overarching issue is just being competitive because the usage of Google AI will be driven largely by its customer base. Many/most who use Google Docs and Workspace, or whatever, will stick with Gemini because it's easier and more seamless, works everywhere they work, and is a single bill.

Microsoft can and will benefit from this same dynamic with Microsoft 365. You can call it inertia, but whatever the name, Microsoft is dominant with the enterprise, and as long as Copilot isn't a complete laughingstock, some percentage of that user base will stick with it. Granted, that's been a small number so far. But there...