Google has been quietly, then very loudly, reorganizing its engineering efforts around one clear priority: getting better at code. What started as an internal experiment has turned into a full-scale push to build AI tools that write, fix, and evolve software.
Google has been quietly, then very loudly, reorganizing its engineering efforts around one clear priority: getting better at code. What started as an internal experiment has turned into a full-scale push across multiple teams to build, refine, and deploy AI tools that can write, fix, and evolve software with minimal human intervention. The numbers coming out of Alphabet now speak for themselves, and the industry is paying close attention.
Alphabet's chief financial officer confirmed during the company's Q4 2025 earnings call that roughly half of all code at the company is now being generated by AI agents before being reviewed by its own engineers. That figure represented a dramatic jump from the 25 percent that CEO Sundar Pichai had disclosed less than a year earlier, and it signals how fast Google is moving to operationalize AI-driven development across its own organization. The pace of change internally is being matched by an equally aggressive push externally, with a cluster of specialized teams racing to build the next generation of coding tools.
Central to that effort is Jules, Google's autonomous coding agent that works asynchronously inside a secure cloud virtual machine, reads the full context of an existing project, and performs tasks such as fixing bugs and writing tests without requiring a developer to stay active in the session. During its public beta period, thousands of developers submitted tens of thousands of tasks to the tool, resulting in over 140,000 code improvements shared publicly. Google has not treated Jules as just a consumer-facing product. The company has already tested it on internal coding tasks and is now making a deliberate push to use it across far more projects within the organization. The team behind Jules has been iterating aggressively since launch. Recent updates included the release of Jules Tools, a command-line interface that allows developers to assign and verify tasks directly from the terminal, reducing the friction of jumping between browser tabs and repository interfaces. Additional capabilities rolled out in quick succession, including memory features that allow the agent to retain user preferences and feedback across sessions, a revised diff viewer, image upload support, and the ability to engage with pull request comments. These are not incremental polish updates. They reflect a team moving with real urgency.
Signals from inside Google now point to a next-generation project operating under the internal name Jitro, an evolution of Jules that moves beyond the prompt-and-execute model that currently defines most AI coding agents. Rather than requiring developers to manually instruct the tool on each task, Jitro appears designed around high-level goal-setting, where the agent autonomously identifies what needs to change in a codebase to move a specific metric in the right direction. Developers would define desired outcomes such as better test coverage or lower error rates, and the agent would work out how to get there. That is a meaningful conceptual shift, from a tool that executes instructions to one that pursues defined goals over time. On the research side, Google DeepMind released AlphaEvolve, a coding-focused agent that uses the Gemini 2.0 model family to generate candidate solutions to algorithmic problems, scores them iteratively, and discards weaker outputs until it converges on the most efficient approach. The tool has already been applied to Google's own internal infrastructure, finding a more efficient way to allocate computing jobs across the company's vast server network. In benchmarks involving matrix multiplication, a fundamental problem in computing, AlphaEvolve improved on records that prior AI tools had only recently broken themselves.
A major industry survey from 2025 drawing on nearly 5,000 technology professionals found that AI adoption in software development had become near-universal, with 90 percent of respondents saying they use it at work and over 80 percent reporting productivity gains. The same research found a critical caveat though: teams with weak automated testing and slow feedback loops were not seeing the same benefits, suggesting that the quality of the surrounding engineering culture matters as much as the AI tools themselves. That context matters for how Google frames its own internal productivity gains, since its engineering organization has decades of infrastructure investment underpinning the AI layer sitting on top. The competitive environment driving all of this is fierce. As of mid-2025, around 30 percent of new code at both Google and Microsoft was being AI-generated, while Meta expected AI to account for half of its new software development within the following year. By early 2026, Google had already cleared that threshold, underscoring how quickly the internal adoption curve moved once the tooling matured. The pressure to maintain a leadership position in coding AI is now both a product and an infrastructure imperative for the company.
Google's intensified focus on coding AI arrives at a moment of significant investment and readiness. The potential economic impact of AI across MENA has been estimated at 320 billion dollars by 2030, with the UAE and Saudi Arabia expected to see the fastest annual growth rates. Google has partnered with Saudi Arabia's Public Investment Fund to expand its Dammam cloud region into a dedicated AI hub, with a specific focus on Arabic-language model development and local digital skills training. That infrastructure investment makes the Gulf a meaningful market not just for AI consumption but for localized AI development. New Arabic-language Gemini features, including coding support, have been rolled out for users across the region, connecting everyday developers and professionals to the same underlying model improvements Google is driving at scale. The UAE government separately partnered with Google to give university students free access to Gemini Pro, a move that seeds the next generation of engineers in the region with direct experience of the tools Google is betting its own development on. As the gap narrows between where Google's internal teams operate and where its public products land, developers across the Middle East stand to benefit from the same productivity shift already reshaping engineering at the world's largest tech company.
