Google's upcoming Gemini Pro is several months delayed due to coding functionalities not meeting internal targets.

      **TL;DR** Google is significantly delayed in rolling out its Gemini Pro upgrade due to insufficient coding performance, leading to discontent among engineers who are moving to Anthropic.

      Google is behind schedule in launching its latest AI model, Gemini Pro, as internal coding targets have not been met, according to a Bloomberg report citing ten current and former employees. The upgrade was anticipated to debut at the May developer conference, but the company has struggled to match the pace of Anthropic and OpenAI, both of which have introduced models that exceed Google's current capabilities in coding. Following this news, Alphabet's shares fell by over three percent.

      Recently, Google revised the training data for Gemini in hopes of enhancing its coding skills, but the outcome was underwhelming, based on input from one of Bloomberg's sources. OpenAI and Meta have also introduced new models that leave Google's AI behind in terms of coding capabilities, increasing the pressure on a team already finding it hard to deliver. A spokesperson for Google stated that the company is "shipping quickly across a wide range of models" and is currently testing the upgraded Pro model, a new Flash model, and other models in collaboration with partners.

      A significant part of the issue stems from organizational structure. Multiple teams—including Google Cloud, DeepMind, and Android—are developing AI coding tools for developers, which has led to internal competition that hampers progress. Co-founder Sergey Brin has advocated for faster advancements in AI coding, but his efforts have been hindered by rival factions within the company and engineers who believe essential coding should still be performed by humans to align with Google's standards, according to former staff.

      Google has begun to streamline its disjointed coding initiatives. Chief AI Architect Koray Kavukcuoglu is making efforts to centralize the company's AI coding tools, and a new team at DeepMind, led by research engineer Sebastian Borgeaud, has been established specifically to address the issue. At its latest Cloud conference, Google announced that 75 percent of its code is now generated by AI and that it has centralized most developer tools under the Antigravity platform, which manages data, memory, and safety protocols for AI applications.

      These delays have led to a wave of senior talent leaving for Anthropic and other laboratories, with former employees citing frustration over Google’s competitive stance as a key motivator. Engineers attempting to utilize AI for their own projects often face capacity limitations due to internal competition for computing resources, a challenge that also impacts external clients. Accessibility to Anthropic’s Claude is restricted within Google, allowing only certain teams engaged in cutting-edge research to use it.

      Customers awaiting the Pro upgrade have experienced varied outcomes with the current Flash model. Figma’s product manager, Rodrigo Davies, mentioned that the model achieves a "sweet spot of speed and quality" for the AI assistant within their design platform. In contrast, Freddy Vega, CEO of the Latin American education platform Platzi, noted that the Flash model is more costly and slower than its predecessor while still being less capable than competitors, prompting his team to transition to Anthropic instead.