What's new with data agents

Key Concepts

• Agentic AI: AI systems capable of autonomous, proactive decision-making and task execution.
• BigQuery Assistant: An AI-powered tool within BigQuery Studio for SQL authoring, troubleshooting, and optimization.
• Data Agents: Persona-specific agents (Engineering, Science, Conversational Analytics) designed to automate complex data workflows.
• MCP (Model Context Protocol): A framework/API standard for connecting AI agents to external data sources and tools.
• Conversational Analytics: A capability allowing users to query data using natural language, including advanced features like forecasting and anomaly detection.
• Deep Data Research (Deep Dive): A feature that synthesizes structured data, unstructured data, and real-time web information into comprehensive reports.
• Agentic Workflows: The ability to configure agents to trigger tasks based on events or schedules.
• Google Cloud Data Agent Kit: A developer-focused toolkit (VS Code extension and CLI plugins) for building and managing custom agents.

---

1. The Agentic Era and Google’s Strategy

Google positions itself at the forefront of the "agentic era," citing Gartner’s prediction that by 2027, over 50% of business decisions will be augmented or automated by agentic AI. Google’s strategy spans the entire stack, from infrastructure to application-layer agents. Their investments are categorized into three pillars:

• Assistive Capabilities: Tools like BigQuery Assistant to boost individual productivity.
• Persona-Specific Agents: Task-oriented agents for data engineers, scientists, and business users.
• Developer Tools: SDKs and MCP APIs for building custom agentic solutions.

2. Persona-Specific Data Agents

• Data Engineering Agent: Automates the creation of complex data pipelines using natural language. It allows engineers to review and modify plans, ensuring control while accelerating development. Now Generally Available (GA).
• Data Science Agent: Operates within Colab notebooks to automate data exploration, featurization, and model building. It significantly reduces the time required for model validation and accuracy testing. Now GA.
• Conversational Analytics Agents: Designed for business users to democratize data access. Available in Looker and BigQuery, these agents support multi-modal data processing and advanced analytics (forecasting, key driver analysis).

3. Advanced Features and Developer Tools

• Deep Data Research (Deep Dive): Synthesizes information from multiple sources (BigQuery, unstructured data, web) to generate long-form, comprehensive reports rather than simple Q&A.
• Conversational Analytics API: Enables developers to embed conversational data capabilities into custom applications.
• BigQuery MCP Server: A fully managed service that eliminates the need for custom glue code to connect BigQuery to open-source frameworks.
• Agent Analytics Plugin: Provides real-time monitoring of agent performance, including latency, token usage, and customer sentiment, with data persisted directly in BigQuery.
• Google Cloud Data Agent Kit: A new toolkit (VS Code extension and CLI plugins) that simplifies the developer workflow by providing a unified interface for catalog viewing, pipeline creation, and model comparison.

4. Case Study: Virgin Media O2 (VMO2)

Mauro Flores (AVP at VMO2) shared the company's journey in implementing data agents to manage 45 million connections.

• Strategy: Centralized data on Google Cloud Platform (GCP) with BigQuery as the "single source of truth."
• Implementation: They built a custom internal tool called "Parity," powered by Gemini Enterprise and the Analytics API.
• Key Outcomes:
  • Reduced agent creation time from months to minutes.
  • Implemented "Golden Queries" to tune model accuracy for critical business questions.
  • Mandatory training for all users to ensure safe and governed use of AI.
  • Impact: 12,000+ business questions answered via the agent, with plans to scale to 5,000 licenses by year-end.

5. Notable Quotes

• "By 2027, more than 50% of business decisions are going to be augmented or automated by the agent AI." — Ganesh Geetha (citing Gartner).
• "Rubbish in, rubbish out. Our strategy has been BigQuery is the single source of truth and we build business layers." — Mauro Flores.
• "This is the best tool I've ever seen built in VMO2." — Feedback from a 20-year telco veteran regarding their internal agent.

6. Synthesis

The transition to agentic AI in data management is moving from simple assistive coding to autonomous, multi-step workflows. By providing both "out-of-the-box" agents for business users and robust APIs/toolkits for developers, Google is enabling organizations to move from static dashboards to dynamic, conversational data exploration. The success of VMO2 highlights that the primary challenge is not just the AI technology, but the underlying data governance and the creation of trusted "business layers" that ensure the AI provides consistent, accurate, and actionable insights.