Comment by devinplatt

It's so interesting to read these comments because this this is literally my job (to help Workspace teams integrate Gemini faster)

Some thoughts:

- Hardware constraints are real, even at Google.

- Features are often released for enterprise users first, before being released for the general public.

- I only started my current role last year.

- 2 years ago is forever ago in AI and things are changing a lot. For example, with older model generations people used to do a lot more fine-tuning (slow) vs. prompt engineering (fast). The implication is that things that are easier to do today might have been hard to do not that long ago. The rapid changes also create churn for internal platforms and dev tooling.

- Google is less yolo and cares about safety, prompt injection, etc. so some time goes into that

- Typical big company bureaucracy also applies, but TBH there's a lot of pressure to deliver Gemini related stuff so I think there's less of that