Google CEO Sundar Pichai has conceded that his firm is trailing rivals in agentic coding, the one AI functionality that has changed into the business’s most profitable battleground. Talking on the Arduous Fork podcast, revealed by The New York Instances, Pichai stated Google’s fashions sit on the frontier on textual content, multimodality, voice and reasoning, however not on the long-horizon coding duties that Anthropic’s Claude Code and OpenAI’s Codex have made their calling card. “In the case of agentic coding with software use, instruction following and long-horizon duties, I feel we’re a bit behind at this second,” he stated.His clarification was blunt, and it was about product, not mannequin high quality. Google by no means had the place the place builders really work, so it by no means bought the information flowing again. “We perhaps did not fairly have the floor, like Claude Code for instance,” Pichai stated, including that he’s “very, very optimistic and assured we are going to push by means of there.” The admission lands awkwardly. Pichai maintains Google is the one giant firm genuinely on the frontier, but the frontier is being outlined by two startups that picked one drawback and refused to let go of it.
Why agentic coding turned the AI business’s most beneficial battleground
Coding is the place the cash went. Anthropic’s income exploded on the again of Claude Code, and OpenAI pivoted from shopper to enterprise with Codex, with one govt reportedly telling employees to cease chasing aspect quests. Wired flagged the form of this combat again in 2023, arguing the true contest was by no means over chatbot parlour methods however over which firm may hook builders first, citing Microsoft analysis that discovered builders work over 50 p.c quicker with an AI assistant. Mordor Intelligence now expects the AI code instruments market to develop from $9.3 billion this yr to roughly $30 billion by 2031.Pichai rejected the suggestion that Google’s unfold of bets is the issue. “I do not see it as a difficulty of focus,” he informed Arduous Fork. “We’re a big firm, and we now have scale, so we will concentrate on a number of issues on the identical time.” He described the tempo of the sphere in phrases that learn much less like reassurance than warning: “30 to 60 days appears to be like like 5 years.”
Antigravity 2.0 and Gemini 3.5 Flash: Google’s reply to Claude Code
Google’s repair arrived at I/O 2026. Antigravity 2.0 shipped as a standalone desktop app with a CLI, an SDK and the power to orchestrate a number of brokers without delay. Gemini 3.5 Flash, co-developed with Antigravity, is pitched as 4x quicker than rival frontier fashions, with an optimised variant Google claims hits 12x. Onstage, engineer Varun Mohan confirmed brokers spawning off to construct separate parts of an working system from scratch, a process Pichai stated would take an individual hundreds of hours.He additionally acknowledged the rocky launch. Tightened utilization limits triggered developer complaints, and Google has since reset the quota. “That’s rightfully a supply of frustration once you encounter it,” he stated. “I really feel the identical.”
Google’s AI coding hole: what the inner numbers really say
Pichai’s proof that the hole is closing is inside adoption. Token utilization inside Google is doubling each week, he stated, and he has “by no means seen something prefer it” within the firm. Getting Antigravity into real-world use, and pulling that information again in, is what he expects to maneuver the needle.The inner numbers reduce each methods, although. Google’s personal CFO has stated Anthropic writes near one hundred pc of its code with AI help, whereas Google sits at roughly 50 p.c. The AI coding strike workforce assembled in April, with Sergey Brin and DeepMind CTO Koray Kavukcuoglu concerned, is already being reorganised. In an inside memo, Brin wrote that Google should “urgently bridge the hole in agentic execution” and switch its fashions into main builders of ultimate code.Worth is the opposite lever. Google launched an AI Extremely tier at $100 a month at I/O and reduce its high plan from $250 to $200, positioning itself because the cheaper possibility for heavy coding workloads. That solely works if the software holds up.