What if OpenAI actually does have a moat?
My bias is to think that the most useful AI is the one that has your personal context — your documents, your emails, and all that stuff.
But despite the fact that ~100% of my useful documents are with Apple or Google, I almost never find myself using Apple AI or Gemini, and instead copying and pasting into Claude or ChatGPT. I have meticulously organized project folders in Google Drive, yet I’m never asking Gemini about them.
Does this mean:
My intuition is wrong - model quality is actually more important; over time, our data will move to where the models are. Owning the user was less valuable than we thought it was.
My intuition is right, but it’s still too early; Apple and Google are big, but eventually they’ll catch up and I’ll start using them
Apple and Google are failing in the way that companies fail when paradigms shift and new competitors will rise up to take their place
There’s a stickiness in how models work. I’ve developed a feel for how to use Claude / ChatGPT, what they’re good at and when they’ll bite me, and there’s a lock in quality here. I did feel a sense of exhaustion at having to learn DeepSeek r-1 despite being excited to have a new player on the scene.
I’m the problem. I’m not as exploratory of a consumer as I think I am and actually there’s a low friction way to use Gemini for these tasks that I haven’t figured out yet. I don’t think it’s this one because it’s pretty common for me to choose Chat GPT 4o when I want an answer fast vs. Claude when I have a project already set up with my key documents in it vs. o1-pro / Deep Research (or one of the other models now) when I need a higher quality answer
Something else I’m not thinking of?
Here is what Claude thinks:
This is an interesting reflection on AI adoption patterns. I think the stickiness factor plays a significant role - there's a real cognitive investment in learning how to effectively "speak" with a particular AI model, understanding its strengths and quirks, which creates a natural barrier to switching even when alternatives might offer theoretical advantages in terms of data integration. This parallels how humans often stick with familiar tools and workflows even when "better" options exist, because the familiarity itself has concrete value.
2025-02-06