When Anthropic’s Claude gradually widened the gap in coding ability, Google finally stopped being silent. according to《The Information》According to reports, Google has formed an internal “strike team” composed of top researchers and engineers with only one goal: to make Gemini’s programming capabilities catch up with or even surpass its competitors. What’s even more unexpected is that Google co-founder Sergey Brin personally participated in this project and sent a memo to all DeepMind employees, admitting that the company “has fallen behind” in this field and requiring all engineers to mandate the use of internal AI agents to handle complex tasks.
From Gemini to Claude: Google finally admits the gap
The battlefield for AI coding assistants has entered a white-hot stage in 2026. Anthropic’s Claude has become the tool of choice for many developers with its powerful program code understanding and generation capabilities. Even the US National Security Agency is reported to be using Anthropic’s undisclosed Mythos model. At the same time, although Google’s Gemini performs well in multi-modal capabilities and general knowledge question and answer, it has never been able to establish the same leadership position in the field of coding.
According to three people familiar with the matter, DeepMind researchers determined after an internal evaluation that Anthropic’s coding tools have significantly surpassed Google’s Gemini model in terms of capabilities. This conclusion obviously shocked Google’s top executives, including Sergey Brin and Chief Technology Officer Koray Kavukcuoglu, who were also directly involved in the operation of this “commando team.”
Memo from Sergey Brin: “Every engineer must use an AI agent”
Sergey Brin’s wording in the memo was extremely direct. He clearly required: “Every Gemini engineer must use Google’s internal agent to handle complex multi-step tasks.” This also revealed several important messages: First, Google has developed an AI agent internally that can perform multi-step tasks; second, Brin believes that forced use is the only way to shorten the learning curve and quickly accumulate data and feedback.
Brin further wrote in the memo: “In order to win the final sprint, we must urgently close the gap in agent execution capabilities and transform our model into the main developer.” This paragraph can almost be regarded as Google’s internal “mobilization order”, announcing a comprehensive investment in improving coding capabilities from top to bottom.
Strike Team’s Secret Weapon: Google’s Internal Code Library
What makes this strike team special is that they have access to a unique resource that no other competitor has access to: Google’s vast internal code library. Currently, Google is training new models directly using internal code libraries to let AI learn to handle truly complex long-term engineering tasks. The advantage of this training method is that the model can be exposed to the most challenging real-world software engineering problems, not just public open source projects.
Google’s internal code library is among the largest in the world in scale and complexity, covering billions of lines of code for search engines, Android, Chrome, YouTube, advertising systems, etc. If Google can really convert these “internal martial arts secrets” into training data, Gemini’s coding capabilities will have the opportunity to grow by leaps and bounds.
Why is Google so anxious now?
Google chose to take a big counterattack at this time. There are several strategic considerations behind it. First of all, coding capabilities have become the core battlefield for AI model competition. Although the developer community is relatively small, it is extremely influential, and their choices often determine the adoption direction of enterprise-level AI tools. Secondly, AI code assistants are evolving from “auxiliary tools” to “agent programs” and can independently complete the complete process from demand analysis, architecture design, code writing to test deployment. Whoever can take the lead in mastering this “end-to-end” code agency capability will be able to take advantage of the future enterprise market.
In addition, Google also hopes to automate more internal development work and even eventually automate AI research itself by improving coding capabilities. If Google allows AI models to write, optimize and train new AI models themselves, this will also completely change the speed and scale of AI development.
In the current AI code assistant market, tools such as Anthropic’s Claude, OpenAI’s GPT-4o and o3-mini, as well as Cursor and GitHub Copilot each have their own strengths. Google’s advantage lies in its huge internal data and computing resources, but its disadvantage is that its productization speed and user experience are often not as flexible as those of start-ups. This time Brin personally took part in the battle, which to some extent reflects that Google’s senior management has attached great importance to this issue to the highest level.
Google I/O 2026: Possible reveal moment
It is widely expected that Google may announce relevant progress at the upcoming Google I/O 2026 developer conference. The annual event, expected to be held in May, will be an important stage for Google to showcase its AI capabilities. If Gemini 3.5 Pro is indeed “will be very powerful” as rumored on the Internet, then Google I/O is likely to be a turning point for Google to regain its voice in the field of programming code.
Gemini 3.5 Pro is going to be insane
Google created a new DeepMind strike team led by Sergey Brin to improve coding models
They are training it directly on their internal codebase to handle complex long term engineering tasks
If they fix the coding then they will surpass GPT… pic.twitter.com/RApGyEHcEa
— Harshith (@HarshithLucky3) April 20, 2026
However, some analysts have reservations. The code-related features Google has launched at its past several I/O conferences, while technically impressive, have not been as effective as expected in terms of actual developer adoption. Whether this “commando” strategy can actually translate into a leadership position in the market remains to be seen.
Summary: The battle for code has just begun
The establishment of Sergey Brin’s commando team marked Google’s official declaration of war on Anthropic. This is not only a competition between the two companies, but also represents a difference in the development route of AI: whether to make AI a “right-hand assistant” for developers, or whether to make AI a “main developer” that can take charge of itself. This competition is undoubtedly good news for developers around the world. Whether Google, Anthropic or other competitors win in the end, more powerful AI coding tools will accelerate the efficiency and innovation of software development. The fact that Brin personally participated in the battle already illustrates the importance of this battle. It is not the first time in the history of Google that a co-founder has directly intervened in product development (the red alert was raised when ChatGPT was released before, and the founder also returned to the company). However, such an open and urgent admission of lagging behind and demanding full mobilization is indeed a rare sight in recent years, and it also makes people look forward to what new things Google will come up with next.
Source: KOCPC Chinese