In my last post, I ended with a cliffhanger: I’m not switching our agency to Google just yet, but I am expanding my testing to see if Gemini has finally overtaken ChatGPT in every way.
But after writing that article, my brain got stuck on a different question. Why did the company that invented the Transformer architecture that made LLMs possible spend so long getting beaten at its own game?
Was it bureaucracy? Was it a "Kodak moment"?
I dug into the timeline of Demis Hassabis, the brain behind Google DeepMind, and I realized I had it all wrong. It wasn't that they couldn't ship a decent chatbot. It’s that they were a little busy playing 4D chess with biology while the rest of us were playing Checkers.
While we were all furiously prompting LLMs to generate '30 days of viral content' in 30 seconds, Hassabis and his team were heads-down on a 50-year-old grand challenge: The Protein Folding Problem.
They weren't trying to optimize email drafts; they were trying to solve medicine.
First with AlphaFold 2, which was a massive scientific revolution. But then came AlphaFold 3, which frankly made AlphaFold 2 look like a kid’s drawing hung on the refrigerator. They were busy winning a Nobel Prize (literally, in 2024), not winning the daily "trending on X" war.
The "lag" wasn't incompetence. It was a trade-off. But when the consumer AI race started threatening Google's search dominance, Hassabis proved something terrifying about focus.
He essentially hit "pause" on the pure science. He merged the teams. He took the minds that were simulating molecular physics and pointed them at a commercial product.
The result was Gemini 3.
As a Head of AI, this reframes my current hypothesis test. We usually judge vendors by who has the shinier demo today. But when you are betting your company’s infrastructure on a partner for the next decade, you have to look at their character sheet.
Do you want the partner whose primary instinct is to ship viral products fast just so we can doom-scroll through endless videos of Harry Potter characters as 1980s bodybuilders? Or do you want the partner who proved they have the sound judgment, the correct focus, and the team to solve the hardest scientific problems on earth?
Now, to be clear: I have nothing against companies driven by financial gain. That is the engine of the market, and it is a perfectly normal, functional motivation. But there is a tangible difference in the output of a team chasing quarterly revenue versus a team chasing a generational impact on human health. When your baseline for "success" is decoding the building blocks of life, that rigor inevitably bleeds into the quality of the commercial products you build. It creates a depth of engineering that "move fast and break things" simply cannot replicate.
So, where does that leave us?
I’m not ripping out our company's current LLM infrastructure today solely because of these findings. We don't make infrastructure decisions based on narrative; we make them based on results. However, my motivation to make that switch has increased tremendously.
We are going to immediately run our tests, evaluate the impact of Gemini 3 across everything we do and make a quick, informed decision.