Google AI Updates: November 21, 2026
Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled
1. Google Research Traces Duplex From Phone-Call Bot to the Gemini Stack That Now Calls Businesses in Tasks
Google Research published a post saying Duplex, the phone-calling AI first introduced in 2018, now powers calls to businesses in Tasks in Gemini. The post traces three architectural eras: closed-domain recurrent neural networks, hybrid stacks using generative models such as LaMDA, and a unified stack built on Gemini. It names sub-second latency over noisy, compressed phone lines and grounding calls in past conversation history as key technical challenges. It also notes that business information gathered from Duplex calls has appeared in Search and Maps over one trillion times, a figure shared in 2023. Source
2. Managed Airflow Gets Agents That Investigate Failed Tasks and a Generally Available MCP Server
Google Cloud described agentic capabilities for Managed Service for Apache Airflow, formerly Cloud Composer. The Managed Airflow Agent in the console investigates failed tasks, recommends environment improvements and checks DAG readiness for Airflow 3, while the Data Agent Kit in IDEs writes, migrates and deploys DAGs. The post also covers the Managed Airflow MCP server, which is generally available, and the Apache Airflow Common AI provider for model-driven tasks inside DAGs. The post does not state a preview or GA status for the console agent or the Data Agent Kit. Source
3. Managed Service for Apache Spark 4.0 Reaches General Availability With Agentic Troubleshooting
Google Cloud announced that Managed Service for Apache Spark 4.0 is generally available for serverless and managed cluster deployments in all regions. The release standardizes on Apache Spark 4.2, Java 25, Python 3.13 and Apache Iceberg 1.12, and adds a redesigned code-first console, faster serverless startups and Lightning Engine acceleration at no extra cost. AI features include agentic troubleshooting that diagnoses failed jobs, and the Data Agent Kit, which configures MCP servers and agent skills so agents can build and run Spark pipelines from VS Code, Claude Code or Antigravity. Source
4. Vodafone Czech Republic Cuts Pipeline Migration Time 90% With BigQuery’s Data Engineering Agent
Google Cloud published a case study in which Vodafone Czech Republic used the Data Engineering Agent in Dataform to migrate legacy, multi-thousand-line BigQuery stored procedures into modular pipelines. The agent parallelized the work and cached intermediate tables with materialized views. Vodafone reports a 90% reduction in migration time compared with manual approaches, pipelines that run twice as fast, and 30% lower costs. Source
5. Google Describes an Internal Data Simulator That Acts as a Wind Tunnel for Testing Enterprise Data Agents
Google Cloud described an internal data simulator for hardening data agents, rather than a launched product. From a short plain-English description, it generates a synthetic company with schemas, rows, event histories, documents and chat threads, plus a benchmark of hard questions with a verified SQL answer key. The pipeline builds a frozen blueprint, simulates tables deterministically from a fixed seed, and checks output for realism and answer-key accuracy. Google says broader multi-agent studies comparing real and synthetic data estates were still underway at publication. Source
6. Gemini for Mac Gets “Computer Use” to Run Desktop Apps, Local Files and Websites in a Background Window
Google added “Computer Use” to the Gemini Mac app, letting Gemini complete multistep tasks across local files, desktop apps and websites. Users describe a goal and Gemini clicks, types and navigates menus and forms in an isolated background window with a live chat preview, so the screen and cursor stay free. Users pick which apps Gemini can access, can watch each step, pause at any time and approve key actions beforehand. Google’s examples include batch-editing photos and 3D scenes, filling in shipping forms from local PDFs and building presentations. The feature rolls out in English in the US. Source
7. Google Rolls Out a Collapsible AI Mode Sidebar and Temporary Threads on Desktop Search
Google is rolling out a collapsible sidebar on the Desktop Search homepage worldwide. In AI Mode it lets users reopen recent AI searches, start and search threads, create notebooks and open settings. Temporary threads, which use no prior context, are not saved to history and do not affect future recommendations, are also launching worldwide. Google Search VP Robby Stein announced the changes on X. Source