The One Thing Google Got Wrong With Its New Gemini AI Model

It has nothing to do with its benchmarks.

Google recently shared an announcement about its new Gemini large language model.

It looks cool. And the benchmarks seem impressive.

But right now, that doesn’t matter. Not to me, and not to thousands of AI developers that are building products.


Because Google got one thing wrong when it comes to releasing an AI platform.

They didn’t actually release the APIs for developers to build with.

So, I can’t just go to a website, get an API key and start trying things out.

That’s supposedly going to come out on December 13th. 7 days from now. Now, some might argue that 7 days is not that long a time. I’d argue that in the AI world, a lot can happen in 7 days. Heck, in just 5 days, we had Sam Altman, the founder/CEO of OpenAI get released and reinstated as CEO.

But even putting the timeframe aside for a second. Right now, at this very moment, Google had my attention, and the attention of thousands of developers. And all those developers might scan the announcement, and look at the benchmarks claiming that the model is better than GPT-4 along many dimensions. But, none of that matters to us right now — because we can’t actually use it.

Instead of writing code that hits the new Gemini model, I’m writing a newsletter post lamenting the lack of an API.

The first rule of building a successful platform: Solve for developer delight and minimize Time To Joy.

Developers are impatient. Waiting does not bring us joy.

Disclosure: I’m an investor in OpenAI. Partly, because they understand the value of developer delight.