Meta's in a strange business for that philosophy because, well... 99% of their income is ads. They model and engage users to sell ads.

It's not great dogfood for employees to try.

And their "AI" situation is murky. They've actually use machine learning internally for a long, long time, but the recent rush to try and productize AI more directly is... mixed.

They had a really good open weights LLM division, and built an interesting ecosystem around those "Llama" models. Small/medium businesses helped expand them. Meta employees interacted with other open source projects, too, and posted their own experiments. It was great! And a prime example of "eating your own dog food."

...But that lab had one failed experiment, so Zuckerberg killed the whole thing. As Zuck tends to do.

And now they have some new division which, from my perspective in the tinkerer community, I would bluntly describe as "a clash of Tech Bro egos." It's generous to call experiments like an "AI CEO" as an attempt to test their own product, but it more closely resembles Zuckerberg's pattern of frantically, nervously engaging in something with the nebulous hope it goes viral like Facebook did.