Google CEO Sundar Pichai on the one AI race Google is ‘losing’ to Anthropic and OpenAI, says: We maybe didn’t quite have the…

Google CEO Sundar Pichai has conceded that his company is trailing rivals in agentic coding, the one AI capability that has turned into the industry’s most lucrative battleground. Speaking on the Hard Fork podcast, published by The New York Times, Pichai said Google’s models sit at the frontier on text, multimodality, voice and reasoning, but not on the long-horizon coding tasks that Anthropic’s Claude Code and OpenAI’s Codex have made their calling card. “When it comes to agentic coding with tool use, instruction following and long-horizon tasks, I think we are a bit behind at this moment,” he said.His explanation was blunt, and it was about product, not model quality. Google never had the place where developers actually work, so it never got the data flowing back. “We maybe didn’t quite have the surface, like Claude Code as an example,” Pichai said, adding that he is “very, very optimistic and confident we will push through there.” The admission lands awkwardly. Pichai maintains Google is the only large company genuinely at the frontier, yet the frontier is being defined by two startups that picked one problem and refused to let go of it.

Why agentic coding became the AI industry’s most valuable battleground

Coding is where the money went. Anthropic’s revenue exploded on the back of Claude Code, and OpenAI pivoted from consumer to enterprise with Codex, with one executive reportedly telling staff to stop chasing side quests. Wired flagged the shape of this fight back in 2023, arguing the real contest was never over chatbot parlour tricks but over which company could hook developers first, citing Microsoft research that found developers work over 50 percent faster with an AI assistant. Mordor Intelligence now expects the AI code tools market to grow from $9.3 billion this year to roughly $30 billion by 2031.Pichai rejected the suggestion that Google’s spread of bets is the problem. “I don’t see it as an issue of focus,” he told Hard Fork. “We are a large company, and we have scale, so we will be able to focus on multiple things at the same time.” He described the tempo of the field in terms that read less like reassurance than warning: “30 to 60 days looks like five years.”

Antigravity 2.0 and Gemini 3.5 Flash: Google’s answer to Claude Code

Google’s fix arrived at I/O 2026. Antigravity 2.0 shipped as a standalone desktop app with a CLI, an SDK and the ability to orchestrate multiple agents at once. Gemini 3.5 Flash, co-developed with Antigravity, is pitched as 4x faster than rival frontier models, with an optimised variant Google claims hits 12x. Onstage, engineer Varun Mohan showed agents spawning off to build separate components of an operating system from scratch, a task Pichai said would take a person thousands of hours.He also acknowledged the rocky launch. Tightened usage limits triggered developer complaints, and Google has since reset the quota. “That is rightfully a source of frustration when you encounter it,” he said. “I feel the same.”

Google’s AI coding gap: what the internal numbers actually say

Pichai’s evidence that the gap is closing is internal adoption. Token usage inside Google is doubling every week, he said, and he has “never seen anything like it” in the company. Getting Antigravity into real-world use, and pulling that data back in, is what he expects to move the needle.The internal numbers cut both ways, though. Google’s own CFO has said Anthropic writes close to 100 percent of its code with AI assistance, while Google sits at roughly 50 percent. The AI coding strike team assembled in April, with Sergey Brin and DeepMind CTO Koray Kavukcuoglu involved, is already being reorganised. In an internal memo, Brin wrote that Google must “urgently bridge the gap in agentic execution” and turn its models into primary developers of final code.Price is the other lever. Google introduced an AI Ultra tier at $100 a month at I/O and cut its top plan from $250 to $200, positioning itself as the cheaper option for heavy coding workloads. That only works if the tool holds up.