Who Wins The AI Assistant Wars? + Why Didn’t Google Build Muse?

0:00 – 15:25

The assistant war stops being hypothetical

The conversation opens with a small comic warning sign of a bigger shift: MG says his Mac is near meltdown because he is running so many assistants at once. That image matters because it captures where the market suddenly moved. Just weeks earlier, he had written that nobody had nailed the true personal assistant experience, and now the field is crowded enough to stress-test a machine. He traces the spark back to Instinct, a startup that got loud very quickly, especially on VC social feeds. MG is skeptical in a very specific way, not dismissive of the product, but alert to the incentives. He jokes that praise on VC Twitter can be financially motivated, a way to get close to a hot company before a deal happens. But he also gives Instinct real credit: it recognized that modern large language models finally made the old dream of an AI assistant feel plausible. He connects this to Apple’s old Knowledge Navigator concept video from the 1990s, a reminder that the idea is ancient, but the underlying capability was missing. What makes this wave different, he argues, is not some magical new interface alone, but the combination of capable models and a smart starting point. Instinct did not ask users to learn a whole new system. It piggybacked on familiar messaging behavior. You text it like you would text a friend or a human assistant, and it figures out the back-end work. That choice solved two classic product problems. First, the blank page problem: when an AI can answer almost anything, people often freeze because they do not know what to ask. Second, the boil the ocean problem: assistant products try to do too much too early and end up vague. Instinct’s answer was proactive prompting. Instead of waiting for a perfect command, the assistant might say it noticed an email that needs a reply, or a bill due in 10 days, and ask whether it should step in. That is the surprising angle in this section: the breakthrough is not raw intelligence first, but guided momentum. The product becomes useful when it gives you a shove at the right moment. MG also explains why the first killer use case may not be one universal task like booking movie tickets. He now thinks the winner will need to do five to 10 things really well, with different ones resonating for different people. Still, there has to be one shareable moment, the thing you tell a friend about. The strongest early example is money. People love asking these assistants to inspect bills, subscriptions, and credit cards to find savings. That is compelling not because it is futuristic, but because it is concrete: save me money, cancel what I forgot, tell me if I am paying the wrong way.