Using Gemini for In-Place Google Cloud Data Querying

For many South African enterprises, the promise of artificial intelligence has been hampered by a persistent bottleneck: the need to migrate, format, and duplicate vast datasets just to make them legible to a model. When your operational data is siloed across legacy SQL servers in a Johannesburg data centre and newer cloud-native storage buckets, the overhead of moving this information to a separate AI environment often negates the efficiency gains you sought in the first place. The October 2026 update to Gemini Enterprise changes this calculus by introducing native, in-place querying capabilities. Instead of treating your data as a static export, Gemini can now function as an intelligent layer that sits directly atop your existing Google Cloud infrastructure, interpreting schema and executing queries in real-time without moving the underlying files.

The technical shift here is from a model-centric view, where you feed the AI a dataset, to an agent-centric view, where the model is equipped with specific skills to interact with your backend systems. By leveraging the updated Vertex AI connectors, Gemini now understands the specific dialect of your enterprise databases. When a user asks a question about regional logistics performance or inventory turnover, the AI does not just hallucinate an answer based on training data. Instead, it translates the request into a precise machine-readable query, executes it against the cloud source, and synthesises the retrieved row-level data into a coherent response. This process ensures that the answer is always based on the most current state of your business rather than a snapshot taken during the last data pipeline sync.

From a cost perspective, this approach directly addresses the silent tax of cloud computing: egress and storage redundancy. In traditional architectures, businesses often pay twice for their data—once to store it in its primary location and again to host a duplicated version in a machine learning-specific environment. Furthermore, constant data replication introduces latency and increases the surface area for security vulnerabilities. By querying in-place, you maintain a single source of truth within your defined cloud boundary. The cost savings are not merely related to reduced storage bills but also to the significant reduction in technical debt associated with maintaining complex ETL pipelines that break whenever your database schema undergoes a minor update.

Implementing this effectively requires a structured approach to defining AI agent skills. You must clearly map which datasets are accessible to Gemini and define the guardrails that prevent the model from performing unauthorized operations on your production tables. This is where many businesses falter by giving the agent too much latitude or too little context. The goal is to provide the model with a set of well-defined functions that correspond to your most common business questions. When the agent has access to metadata—such as field descriptions, relationship mappings, and privacy constraints—it can navigate your cloud environment with the same precision as a senior data analyst, provided it is directed by the right intent.

Security is paramount when allowing an AI model direct access to your data, and the latest Google Cloud framework incorporates rigorous Identity and Access Management integration. Because the agent queries data under the identity of the user asking the question, you can enforce row-level security and data masking without building custom middle-ware. If a regional manager queries regional sales data, Gemini only accesses the rows that the manager is permitted to see. This makes the implementation compliant with South African data protection regulations like POPIA, as you are not creating new, unsecured copies of sensitive personal information in an external testing environment.

Developing these skills for your specific business case involves more than just enabling a setting in the console. It requires a thoughtful assessment of your current database architecture and the identification of the high-value, high-frequency queries that drive your operations. You need to identify where your data is clean, where it requires transformation, and how Gemini should handle gaps in data. This is an engineering problem as much as a data science one. It involves configuring APIs, testing query performance, and ensuring that the AI agent can handle edge cases without defaulting to generic or inaccurate answers when it encounters ambiguous data structures.

For most operations leads, the challenge is not understanding the value of AI but navigating the complexity of implementation in a way that remains stable and secure. The technology is no longer in the experimental phase, yet the difference between a prototype that produces interesting results and a production-grade system that powers daily decision-making is substantial. You need a partner who understands the nuance of the Google Cloud ecosystem and can build these integrations so they perform under the load of a real-world enterprise. At WriteNow Agency, we specialise in translating these advanced cloud capabilities into practical, reliable tools that work for your team from day one. If you are looking to integrate in-place querying into your existing operations, let us start a conversation about how to tailor these AI skills to your specific data environment.