onebit0fme

I think this is mostly right. But there’s something missing here: agentic tuning. Lately, OpenAI and Anthropic have focused on models that are well trained to use tools, not just to do well on standard benchmarks like GPQA or MMLU. You’re probably right that the most connected platform will come out ahead. But they’ll need to keep their models sharp at using these tools. And with more and more MCP tools, we might end up with context windows packed full of just tool calls. That could be a problem.

Kinda fun, it'd be more fun to explore any street view

> my current hypothesis: the return trigger isn't gamification, it's social - knowing someone else is learning the same thing, or that someone will notice if you stop

My honest take is that it's likely both. Increased social stakes is a big one, but it's incredibly difficult to build. It's hard to convince someone to use your app consistently, it's even more difficult to convince 2+ to keep each other accountable. My point is that it might be difficult to test in isolation, or even lead with this as the primary strategy. What you should achieve first is Sean Ellis test (something my friend from product taught me once) - answering single question “How would you feel if you could no longer use this product?” When this test holds (roughly >= 40% answer they'd be very disappointed), then it enables sharing with others, which enables social components you're describing.