DeepMind engineer on AI tool usage

A Google Cloud artificial intelligence executive said more than 40,000 software engineers at Google use “agentic coding” tools every week, pushing back on claims that production use is still sparse. (venturebeat.com) The figure came from Addy Osmani, a director working on Gemini, Vertex, and agents at Google, in a public exchange on X on April 14, 2026. In the same discussion, Osmani said Googlers use custom models, command-line tools, and model context protocol systems, and can also access Anthropic models through Vertex. (venturebeat.com; addyosmani.com; talks.addy.ie) “Agentic” tools are software systems that do more than autocomplete a line of code. They can take a task, break it into steps, call other tools, write or edit files, run tests, and return with a draft result for a human engineer to review. (gartner.com) The exchange landed in a broader argument over whether large companies have actually moved artificial intelligence from demos into daily engineering work. Google’s own 2025 DevOps Research and Assessment report said 90% of surveyed technology workers use artificial intelligence at work, based on nearly 5,000 respondents and more than 100 hours of qualitative research. (research.google; cloud.google.com) That same report said artificial intelligence acts more like an amplifier than a fix: strong teams get faster, while weak testing, slow feedback loops, and tightly coupled systems can turn added speed into instability. Google Cloud said the 2025 data showed a positive relationship between artificial intelligence adoption and software delivery throughput, while stability still lagged. (cloud.google.com) Outside Google, survey data points in the same direction but with more caution. Stack Overflow’s 2024 developer survey found 76% of respondents were using or planning to use artificial intelligence tools in development, 81% cited productivity as the main benefit, and 31% said they were skeptical of output accuracy. (survey.stackoverflow.co) Among professional developers, workplace access still looked uneven in 2024. Stack Overflow found 32.4% said their company had artificial intelligence-assisted technology available at work, up from 15.7% a year earlier, and only about half said they had tools to quickly understand any area of their company’s code or systems. (survey.stackoverflow.co) The scale of Osmani’s number is notable inside a company Alphabet said had 190,820 employees as of December 31, 2025, though that total includes many workers who are not software engineers. The post did not define exactly how Google counts a weekly user, what qualifies as an agentic tool, or how many of those users rely on the systems for core production work rather than occasional tasks. (sec.gov) What the public exchange showed, more than anything, is that the argument has shifted from whether engineers can access these tools to how deeply they are woven into real software delivery. Google’s own research puts the same question in plainer terms: adoption is widespread, but results still depend on the team using it. (venturebeat.com; cloud.google.com)