Google’s Gemini AI Researcher Shazeer Joins OpenAI

Noam Shazeer has left Google for OpenAI. Maybe you haven’t heard of him, but you should know his work. He was a co-author of the 2017 paper “Attention Is All You Need,” which presented the Transformer architecture used by virtually all large language models today (including ChatGPT, Gemini, Claude, etc.). He’s also been a Vice President of Engineering at Google and co-lead of the Gemini models with Jeff Dean and Oriol Vinyals. “This was a hard decision to leave,” he said about his departure.

Why should you care? Because his career tells us why this matters. He started at Google in 2000, building search features like spell checking before departing in 2021 to co-found Character.AI, an early conversational AI company. In 2024, he came back to Google in a ~$2.7 billion acquisition that reunited him and his co-founder Daniel De Freitas to help improve their reasoning capabilities. Now he’s gone again, joining OpenAI in an unspecified capacity — the second major AI personnel shift of 2026 after Andrej Karpathy went to Anthropic earlier this year.

1Why one hire counts as news

It might seem odd to make such a big deal out of someone switching jobs. But there’s a good reason: advances in AI happen at the margins, where a tiny group of people understand how to make these systems work better. And where they work is a strong predictor of which products will improve over the next couple of years. So when someone involved in creating the foundational architecture switches labs to their main rival, it sends a message about momentum — and competition among these firms for top talent.

What does this mean for you? Well, first off, remember that your favorite tools are developed by particular individuals rather than some abstract “AI.” Second, the difference between vendors is small enough that a single person can make a big difference. Personally, I wouldn’t put too much stock in individual moves — there’s always turnover at the top — but pay attention to trends. If lots of senior folks are going to a particular lab, expect to see improvements in their products within six to twelve months.

In general, 2026 has seen a shift towards focusing on model creators as well as models. This reflects just how important this field has become — and serves as a reminder that the competition isn’t over yet, and it’s being decided not only by who has the largest datacenter but also by who’s sitting around the table.