Google has spent several months away from the frontier in AI, and it might’ve been because of not focusing enough on models with technical skills.
Google’s Logan Kilpatrick says the company should probably have put more resources into coding sooner. The admission comes as the industry has become “very code pilled, very science pilled,” in his words, and Google finds itself playing catch-up.
“I think we were like a little, it’s not that we were late to the game because folks knew it was important, but in hindsight everything is much more clear,” Kilpatrick said on a podcast. “I think in hindsight now is obvious, like we should have, from an order of magnitude of resource allocation, probably put more into coding sooner.”
Kilpatrick was careful not to call the alternative a mistake. “We had a bunch of other stuff that we were doing which those things actually turned out quite well,” he said, pointing to Nano Banana as the prime example. “A great incredible image model that sort of took the world by storm, had this massive impact for our consumer products and a bunch of other parts of the business.”
He acknowledged the tradeoff. “To make that model, it took research and compute and time, and had this huge impact. And was it in hindsight right to do that versus doing something on coding? I don’t know that there’s actually a clear right answer.”
A genuine hit
Kilpatrick’s defence of Nano Banana holds up. Traffic to Google AI Studio jumped from around 3 million daily visits to around 4.5 million right after the model’s August 26 release, and a Google executive said it had brought 10 million new users to Gemini. Within weeks, Gemini became the top app on the App Store, helped along by viral trends like AI figurines. The model had first appeared anonymously on LMArena, where its editing quality drew attention before Google hinted that it was behind it.
Google kept building on it. Nano Banana Pro, launched in November 2025, crossed 1 billion images in the Gemini app in 53 days, and Nano Banana 2 later became the default across the Gemini app, Search and Flow. Few AI companies have had a consumer moment like it.
The coding gap
Meanwhile, the money in AI has moved toward code. Anthropic says its annualized revenue run rate has touched $47 billion, driven by enterprise adoption, and a significant share of that growth is attributed to Claude Code. The Information has reported that Anthropic is now generating 35% more revenue than OpenAI, with Claude Code a breakout product behind it.
Google’s coding tools have had a harder time. A recent industry analysis described the company as a significant player that appears to lag in this niche, while Google retired free and individual access to Gemini CLI in June in favour of a successor tool, Antigravity CLI. That is the kind of gap Kilpatrick seems to be describing.
There have been bright spots. Google’s Flash models have been closing in on rivals on coding benchmarks. Gemini 3.7 Flash led Code Arena’s WebDev ranking and edged out Claude Sonnet 5 and GPT-5.6 Terra on FrontierCode, and it went on to become Google’s fastest-growing model ever. Kilpatrick has also said that vibe coding started really working after Gemini 3.
Beyond coding
The problem is broader than one use case. Coding is the most visible front in a race that is increasingly about raw intelligence, and Google is not leading it. Gemini 3.8 Flash scores 59 on the Artificial Analysis Intelligence Index, behind Claude Fable 5.1 at 66 and Claude Opus 5 at 63, and also behind GPT-5.6 Sol at 61.
There is also a gap at the top of Google’s own lineup. Google hasn’t shipped a true successor to Gemini 3.1 Pro since February, and OpenAI’s leaked GPT-6 Astra benchmarks suggest the frontier is moving further away still.
Kilpatrick’s comments amount to an admission that the choice of where to spend compute has costs, even when it produces a hit. Nano Banana pulled millions of people into Gemini. But the frontier is being set by the labs that bet early on coding and agents, and Google is now working to close that distance. And with Gemini 4 showing supposedly promising early reactions, Google might be preparing to reclaim the frontier.