Google Gemini has the worst LLM API

Re:Google Gemini has the worst LLM API by djohnston on Sunday May 04, 2025 @09:41AM (#43886657)

Google APIs have always been difficult for me to use. I find their documentation super fragmented, and they always build the same thing twice with awkward overlaps. In this case, it's Vertex and GenAI - two different SDKs with isomorphic data structures that are seemingly incompatible, but only sometimes. I don't understand how these things happen - but as usual I blame PMs trying to mark their territory and pissing all over everything.

The new gemini models themselves though, are killer. The confusion is a small price to pay.

[ Parent | on HN ]

Re:Google Gemini has the worst LLM API by harlysparks on Sunday May 04, 2025 @11:23AM (#43887233)

The duplicative functions and overlapping features is a product of how their perf annual review cycle works and the way you have to play the game to get promotions as a SWE there. I think they've changed the setup in the last few years, but historically Google rewards building lots of beta versions of stuff with some commercial potential, throwing it over a wall and seeing what sticks.

[ Parent | on HN ]

Re:Google Gemini has the worst LLM API by smel on Sunday May 04, 2025 @01:26PM (#43888042)

This happen when you have 200k employees competing internally more than with outside competition.

[ Parent | on HN ]

Re:Google Gemini has the worst LLM API by shresbm123 on Monday May 05, 2025 @01:16AM (#43892115)

Haha as one of those PMs - thank you for appreciating the models. We do have a unified SDK and are working on making the docs more comprehensive and clearer. But do let keep giving us specific and candid feedback

[ Parent | on HN ]